Abstract

Cancer is one of the leading causes of death in the paediatric population. Artificial intelligence (AI) can contribute to the development of medical diagnostics which can lead to early detection and therefore reduce the mortality rate of paediatric cancers. In this scientific paper, we will discuss how we can use artificial intelligence for the early detection and diagnosis of cancer in children. Artificial intelligence (AI) systems can distinguish patterns and recognise anomalies that can indicate the presence of a cancer through a detailed and exhaustive analysis of numerous medical images, genetic information, and patient records. This new innovation can contribute to better accuracy in the diagnosis itself, reduce the workload, and achieve quick results. Furthermore, with the assistance of AI technology, patients would have access to treatment plans adapted for them. Artificial intelligence, using deep learning and recognising different features and patterns, can detect whether it is a malignant cancer or a benign cancer. AI has the potential to reduce errors in diagnosis, help in planning the best choice of treatment, and thereby improve patient outcomes. While AI is useful in more efficient, faster and safer detection of early signs of cancer in children, there are many aspects of contention to consider before its use. These include mandatory professional supervision and patient data privacy. Ultimately, AI has promising prospective success in childhood cancer detection. Reduced possibility of error, quick findings, and a randomisation system are just some of the results of the introduction of modern technology into the world of paediatric oncology. Looking ahead, improvements in AI for paediatric cancer detection will aim to magnify diagnostic accuracy, combine diverse data sources, and ensure strong data privacy. Efforts will focus on personalising tools, reducing biases, and smoothly incorporating AI into clinical practices to better detect and treat cancers early. In this paper we will discuss why artificial intelligence can play a key role in the daily diagnosis of cancer in the paediatric population.

1. Introduction

Paediatric cancer is defined as the group of cancers that affect children, which refers to infants to age 14, and teenagers aged between 15 and 19 (Cleveland Clinic, 2023). Cancer is the uncontrolled division of cells resulting in a tumour, an irregular mass of cells. They can be benign or malignant, in which only malignant leads to cancer. Paediatric cancers are more spontaneous and random in comparison to adult cancers as they lack the accumulation of mutations and interactions with environmental factors that ultimately cause cancer, making it hard to prevent (Dana-Farber Cancer Institute, 2018). Common paediatric cancers include leukaemia, lymphomas and brain cancer (Cleveland Clinic, 2023). Acute lymphocytic leukaemia and acute myeloid leukaemia accounted for 12.4% and 9.4% of paediatric cancer deaths, respectively (Siegel et al., 2020). 

The high death rate of paediatric cancer is indicative of the challenge in diagnosis and treatment. This article will review artificial intelligence (AI) use in paediatric oncology, which could potentially lead to an efficient diagnosing process and effective treatment administration. AI can process diagnosing-related test results more rapidly, which can significantly impact clinical decision-making and is critical in allowing patients to get better quickly. However, as artificial intelligence becomes more common, ethical concerns over its use have arisen. We propose that robust guidelines be put in place to ensure that the sensitive data of young patients is well protected.

2. AI in the Diagnosis of Paediatric Cancer

2.1 Cancer in Paediatric Population

Paediatric cancer is much rarer than cancer in adults. Childhood cancer accounts for only 1% of all diagnosed cancers. In the last 30 years, an increase in cancer cases has been observed in the paediatric population. However, at the same time, the number of deaths due to paediatric cancer has decreased by more than 50% in adolescents and children younger than 20 due to technological advancements (National Cancer Institute, 2024). Children most often suffer from leukaemia, blood cancer, which accounts for 30% of all cases of paediatric cancer. The second most common type of cancer are the cancers of the central nervous system combined with all other solid tumour cancers, which account for 25% of all cases of paediatric cancers. Lymphomas are the third most common, and account for 11% of all cases of paediatric cancers. Soft-tissue sarcomas represent 7% of all paediatric cancer cases. Around 6% of cases are diagnosed as renal cancers or neuroblastomas, along with other cell cancers of the peripheral nervous system. The remaining 16% are diagnosed as other, rarer types of cancer (UK Health Security Agency, 2017). Leukaemia is the most common type of cancer in children. There are two main classifications of leukaemia:

1. Acute lymphoblastic leukaemia (ALL) where the cancer attacks blood and bone marrow. ALL disturbs the normal function of cells, and therefore cells cannot fight cancer.
2. Acute myeloid leukaemia (AML) affects platelets and blood cells.

Lymphoma describes cancer found in a child’s lymphatic system. Lymphoma can be categorised in two groups, which are: Hodgkin and Non-Hodgkin lymphoma. These types of cancer usually are diagnosed in teenagers. 

Another type of cancer that is frequently diagnosed in children is brain cancer. Brain cancer is an example of solid tumours. These tumours are of several types, such as: neuroblastoma, retinoblastoma, osteosarcoma, Wilms tumour, etc. (Cleveland Clinic, “Childhood Cancer,” 2024).

2.2 Diagnosing Childhood Cancer

Detection of cancer in children is demanding. The first reason is that there are no specific symptoms of cancer in the paediatric population, and the symptoms of cancer very often coincide with other illnesses. Some of the symptoms of cancer are fatigue, fever and weight loss, but these symptoms can also be indicators of numerous other diseases (American Cancer Society, 2019). Also, cancer in children is much less common than in adult patients, and means that doctors have less practice in this area, and therefore may not have the necessary experience to recognise cancer in children in good time (Darlington, 2020).

Another issue can be seen in the approach to testing for cancer. Diagnostic approaches are complex, expensive and/or invasive, and, in addition, they often do not show the diagnosis in time or the medical staff may not be concerned by small, hardly visible changes in the image. In this case, a long-term, exhausting, often painful situation becomes necessary, which is stressful for both the patient and their family, as well as for the medical staff. CT (computed tomography), MRI (magnetic resonance imaging), and ultrasound are often used to diagnose cancer in order to see the position and appearance of the mass itself. However, a biopsy is crucial for determining the diagnosis and further treatment, as well as the stage in which the cancer is. Finally, blood and urine are tested to see possible irregularities.

CT is an X-ray that takes pictures from different angles and positions of the body, for the best view possible. It creates detailed 3D images that help to see the exact location and size of the tumour. It is suitable for detecting various cancers, such as kidney cancer. MRI is a medical procedure in which the machine uses radio waves and a magnetic field to gain insight into the inside of the body. MRI is specifically designed to detect conditions that affect soft tissue, such as various tumours. With the help of this method, we can take images of the brain, eye, ear, knee and stomach. Except for leukaemia, most cancers cannot usually be detected through blood tests like a complete blood count (CBC). Even so, certain specialised blood tests can identify tumour markers – chemicals and proteins that may be elevated in the blood when cancer is present.

When cancer is detected, it is important to start treatment as soon as possible. The options used today are chemotherapy, radiation, surgery (when the cancer’s position is suitable for it), and immunotherapy. Often doctors decide to combine  these options, for the best possible therapy. Drugs used in chemotherapy are injected intravenously, attacking fast-growing cells, which is a solution that is undoubtedly suitable for most paediatric cancers. Chemotherapy suits paediatric patients better than adults.

2.3 AI in Diagnosing Paediatric Cancer

Results can take up to two weeks when diagnosing paediatric cancer, which additionally results in anxiety, worry and stress for the patient; we can therefore see that the use of artificial intelligence in this field brings numerous benefits. It has been shown and proven that the addition of artificial intelligence to the practice of diagnostics speeds up the review of recordings by 30%, and thus helps to speed up treatment decisions (Yang et al., 2023). AI can be used to distinguish malignant from benign tumours by studying numerous findings and radiological images, and by analysing the patient’s history and clinical notes. AI improves diagnostic accuracy by 10-15% (Yang et al., 2023). AI will thus have the opportunity, along with the data on the size and type of tumour that it analyses, to help the doctor create a personalised plan for the patient that would ensure the best possible recovery. Although the introduction of AI in radiological assessment would be useful, AI would also serve in the interpretation of histopathological findings. After detailed training, the AI ​​will be able to distinguish between types of cancer and help identify cancer stages. 

Although the introduction of this modern and popular software in the medical industry seems like a hot topic nowadays, we are faced with a few limitations. Data privacy and the mandatory control of experts is something we have to rely on when it comes to this topic. It is certain that artificial intelligence will never be able to replace doctors. The lack of emotional access is extremely important to consider, so technology as such is only an additional help – an added pair of eyes – to ensure the best possible care for the most vulnerable part of society. AI algorithms are revolutionising medical imaging for adult patients by enhancing processes such as image acquisition, reconstruction, quality control, segmentation, analysis, and interpretation. However, the development of AI algorithms specifically for paediatric applications, particularly in paediatric oncology, remains limited. An overview by the American College of Radiology indicates that only 3% of FDA-approved AI applications for medical image analysis (seven out of 221) are designed for paediatric imaging. Furthermore, none of these applications address oncologic conditions in children, underscoring a significant gap in leveraging AI tools for the diagnosis and treatment of paediatric cancer.

Many AI algorithms designed for paediatric cancer applications need to be specifically developed for children. However, the limited prevalence of childhood cancers and the rapid advancement of imaging technologies create challenges in assembling large, standardised datasets, which impedes the effective training of AI models for paediatric oncology. The rarity of childhood cancers results in a smaller patient population, and the associated market size and potential financial returns might not be attractive enough to developers. Moreover, for children with cancer, there is a significant lack of whole-body datasets that are publicly available. As of March 19, 2024, out of 208 imaging datasets available on The Cancer Imaging Archive, only eight are dedicated to paediatric cases. While The Cancer Imaging Archive collaborates with the Childhood Cancer Data Initiative to provide high-quality CT, MRI, and PET images for various paediatric cancers, other public repositories, like the Medical Imaging and Data Resource Center, are still expanding and currently lack a focus on paediatric oncology. The 2017 Radiological Society of North America (RSNA) database offers a large set of paediatric hand radiographs, which can serve as normal reference data (Kumar et al. 2024).

i. Leukaemias

The introduction of AI in order to diagnose the most common type of cancer in children is important for the future of paediatric oncology. A recent study presents data in which AI, in detecting leukaemia in paediatric patients, had an overall accuracy of 88%, sensitivity of 80% and specificity of 90% (Yang et al., 2023). These results focus attention on the significant potential that comes with the introduction of artificial intelligence in the diagnosis of childhood cancer. Increased diagnostic precision, combined with standard and traditional methods, certainly with the presence of a doctor, especially emphasises the importance of this type of technology for a complete revolution regarding leukaemia. Just by introducing AI into routine practice, we can get more reliable results earlier and treatment plans designed for each patient individually.

ii. Abdominal Tumours

Wilms tumour (nephroblastoma) is a kidney cancer that occurs most often in young children. In order to detect this type of cancer early, a support vector machine (SVM) has been developed (Ma et al., 2022). This model was perfected on the preoperative CT findings of 118 children. Accuracy of 79%, sensitivity of 87% and specificity of 69% were shown. However, more precise clinical guidelines are awaited for the further development of such a system for distinguishing all four stages of Wilms tumour. Most importantly, the possibility of using AI technology in diagnosing other paediatric abdominal tumours and non-Wilms tumours is also being investigated. AI has a significantly higher sensitivity for detecting non-Wilms tumours, with 78.1% compared to the pure human one, which is 13-20% (Singh et al., 2024). Judging by the results of this article, the development of an AI model that will better distinguish Wilms tumour from other abdominal tumours, it is concluded that the use of AI technology would reduce the possibility of malpractice.

iii. Musculoskeletal Tumours

In order to predict whether there is a risk of pulmonary metastases in patients with osteosarcoma, a radiometric model based exclusively on machine learning has been built (Pierra et al., 2021). This model was perfected so that the system analysed data from 81 patients. The result is that this way we get an accuracy of 73%. Deep learning is a method in which AI learns to recognise certain patterns in the same way the human brain does. This method was used by Teo et al. when they were developing a model for evaluating chemotherapy response in high-grade osteosarcoma. Multimodal MRI was used in the research in which ten patients of the paediatric population were included. An average accuracy of more than 90% was achieved in distinguishing between necrotic, that is, dead, cells and viable tumour areas only with the help of MRI images. As it was done with the mentioned radiometric method, every system used for diagnostics must be constantly improved, so that there is no error in the system. This kind of mistake is worse than a regular doctor’s mistake, because a mistake caused by a malfunction in the system, without the supervision of an expert, will not only harm one patient, but many of them, until it is discovered that there is a problem. Multimodal MRI is another model on which previous research is based. This model brings improvements in the accuracy of disease prediction, which leads to the fact that it is possible to respond on time and with quality. Therefore, the rates of survivors of osteosarcoma have increased.

iv. Pulmonary Nodules

Lung (pulmonary) nodules are abnormal growths found in the lungs and are mostly benign. AI algorithms show good results regarding the detection of lung nodules, which is promising for future developments. Ni et al. noticed that the deep CNN model of the algorithm mentioned above outperformed the assessment of young medical doctors in detecting nodules in patients with osteosarcoma by a smaller amount (92.3% vs. 90.8%). This model has the advantage that it greatly simplifies the reading of the findings. A study by Hardie et al. found that CAD systems, which are otherwise tested and used for adults, have far less sensitivity for paediatric chest CT. This statement is confirmed so that FlyerScan 68.4% MONA is 53.1% in images of paediatric patients, if we compare it with the results of adult images, which are FlyerScan 83.9% and MONA 95.5%. The cited data from the study show that AI really gives exceptional results in the field of cancer. Nonetheless, it is important to emphasise the fact that it is necessary to carry out special investigations intended for the paediatric population. This conclusion is also confirmed by the marked low sensitivity of the CAD system in paediatric chest CT scanning. FlyerScan and MONA are models that gave notably worse performance in children, if we compare them with the results of the same models, but only in adults. There is a major need to increase paediatric research; the development of AI tools is needed so that they can be accepted in the field of paediatrics as well. Solving this problem would ensure that the youngest patients receive the best possible treatment with accurate diagnostic evaluations.

v. Brain Tumours

Six studies were conducted that investigated the use of artificial intelligence in the detection of brain tumours in children. The result of five out of six of those investigations went in favour of AI. However, clinical applications of AI in the detection of paediatric brain cancer have never been made (Huang et al., 2022). In this study, the disparity between the promising and positive research results and the huge lack of application of the same in clinical situations is visible. It is extremely important that AI starts to be applied in practice if the numerous advantages are considered, while respecting the rules of ethics, which we cover in the third section of this paper. It is essential to solve this situation by conducting research dedicated to the paediatric population, in order to develop more contemporary and modern ways of detecting brain tumours in children.

vi. Lymphomas

As was the case with other types of cancer, research was also done for the use of AI for lymphoma, but in adults. Very few of these studies can be used in the paediatric population. Wang et al. introduced the possibility of detecting lymphoma in children with the help of artificial intelligence software, which would analyse FDG, PET and T1 MRI. The approach to this research was done in two phases, namely 50 data sets for training and 20 for testing purposes. In the end, the multimodal fusion method was successful, solving the problem of false positive results. The implementation of the multimodal fusion method in this study solved a major problem in diagnostics, namely false positive results, which, accordingly, improves the accuracy of paediatric lymphoma detection. In the diagnosis of lymphoma, we are again faced with the problem of a lack of studies dedicated to the paediatric population.

3. AI in Treatment of Paediatric Cancer 

After confirming cancer from lab test results and scans as mentioned above, treatment plans would be put together swiftly. Young cancer patients will receive treatments such as radiotherapy, chemotherapy, surgery, and immunotherapy (Cleveland Clinic, 2023). Treatment options and effectiveness are unique for each patient (Scott, J., 2023); therefore, incorporating the transformative power of artificial intelligence in treatment planning may forecast possible impacts of therapeutic interventions introduced and allow better clinical decision-making, improving efficiency and quality of patient care.

3.1 Chemotherapy Side Effect Predictions Generated by Artificial Intelligence

One of the most common treatments for paediatric cancer is chemotherapy (Yale Medicine, 2024). Differing from adult cases of cancer, paediatric cancer is more aggressive and tends to grow more rapidly (Dana-Farber Cancer Institute, 2018). Most chemotherapy is particularly useful in suppressing fast-growing cancer, and children also exhibit remarkable efficiency in recovery after exposure to high chemotherapy doses, making it a suitable treatment option (American Cancer Society, 2024). Chemotherapy are powerful drugs that can attack cancer cells by interfering with phases of the cancer cell cycle, thereby impeding cell growth (Children with cancer UK, n.d.). For example, alkylating agents are chemotherapeutic drugs that damage the DNA of cells, consequently disrupting cell growth and terminating cancer cell multiplication (Altomara, 2024). 

However, administering these powerful drugs may result in unwanted side effects. Common short-term side effects include fatigue, nausea and hair loss, and long-term side effects include infertility, second cancers and liver, kidney and heart problems (Children with cancer UK, n.d.). Wilms tumour, a common paediatric kidney cancer, requires chemotherapy as part of the treatment. However, it may cause chemotherapy-induced myelosuppression (CIM) to arise. CIM is a severe side effect in which the bone marrow produces insufficient blood cells and platelets, leading to anaemia and lymphopenia (Eldridge, L., 2024). Once they arise, chemotherapy may be delayed as resolving the life-threatening CIM is often prioritised. This reduces the effectiveness and efficiency of the treatment. An online survey was conducted on adult patients who experienced myelosuppression after chemotherapeutic treatment. Results indicated that 88% of participants considered CIM negatively impacted them (Epstein et al., 2020). Therefore, controlling these uncomfortable side effects is essential to providing quality patient care, especially in vulnerable, more sensitive young patients. 

The machine-learning-based prediction model, XGB, was developed by Li and his colleagues to predict the risk of children suffering from CIM if given chemotherapy. The model was integrated into a hospital information system in the form of a clinical decision support system to assess the generalisability in real-world clinical practice. The model would display and rank the variables affecting CIM, with the most critical features first, which are haemoglobin, white blood cells, alkaline phosphatase, coadministration of highly toxic chemotherapy drugs, and albumin. Diminished levels of these features are frequently found in patients with CIM due to blood loss during surgery before chemotherapy. Furthermore, the model calculates a patient risk score and divides patients into low-medium and medium-high risks. 

This clinical model is instrumental in enhancing doctors’ decision-making efficiency and refining the quality of patient care. When this approach is taken in a clinical setting, patients begin by completing haematological and biochemical tests after admission. Armed with the data from test results, the doctor then establishes a chemotherapy treatment and a CIM risk score will be generated by XGB. With XGB, patients with high-risk scores would be identified and alternative treatment plans can be considered, optimising the effectiveness of chemotherapy and ensuring patient well-being (Li et. al., 2023). Chemotherapy’s side effects are particularly accentuated in paediatric cancer as these unwanted side effects not only negatively impact young patients’ physical health, but also create an emotional burden on them and their families. This model can significantly reduce the death rates of paediatric cancer and would be useful if widely applied in hospitals. 

3.2 Chemotherapy Drug Combination Side Effect Prediction by AI

The model above predicts the likelihood of a particular side effect occurring, and algorithms are also developed to ensure the correct drug is chosen, which would also contribute to reducing side effects. Combinations of chemotherapy drugs are often used to mitigate drug resistance and increase the effectiveness of treatment and patients’ quality of life. Drug resistance accounts for most of the chemotherapeutic failures (Król et al., 2010). For example, cancer cells may have increased tolerance to DNA damage and become less responsive to chemotherapy drugs. It is therefore important to provide therapies that heighten the chance of patient survival. Models are developed to assist in the selection of the correct combination. 

A model SyDRa by Li and colleagues was developed to identify synergistic drug combinations (Li et al., 2017). The assistive tool incorporated a feature to assess the resemblance in perturbation effect (drug targeting pathway), chemical structure, and gene expression profile of each drug in the combination. If the drug within the combination possesses a similar chemical structure, it may compete for the same target, thereby diminishing its effectiveness within the combination. The gene profile expressed after treatment administration reflects the patient’s biological response to the medication and is a useful indicator of the effectiveness of the drug. The similarity in gene expression profile is a crucial factor in contributing to drug synergism. Another feature installed seeks distinctions in drugs with the assumption that different drugs can complement each other. The SyDRa model is an effective computational model for anti-cancer synergistic drug combination screening. 

The SyDRa model excluded specific disease pathways during model development. In the future, a more reliable and accurate result can be achieved by involving specific types of paediatric cancer pathways so the drug combination can target the type of cancer specifically. By incorporating this disease-specific information, more personalised and effective cancer treatment can be provided to young patients. 

A recent study has proposed another machine learning approach to forecast the effectiveness of synergistic drug combinations for FDA-approved cancer drugs (El-Hafeez, T.A., 2024). This model resolved the limitations mentioned above by including types of cancer for a more accurate establishment of drug pairs. The data set used was pre-processed, and the model utilises a combination of classification and regression. 

The classification model analyses the pre-processed data and classifies the interaction of two drugs as synergism, antagonism, or additive. Synergism refers to the combined effect of the two drugs being greater than the sum of each individual drug contribution; antagonism is when the combined effect is less than the sum of the individual components; additive arises as the combined effect is the sum of each independent drug (only if linear dose-effect relationships of each drug were present) (Gracía-Fuente et al., 2018). The regression part provides a precise numerical value as the combination sensitivity score (CSS) for the drug combination mechanism. Synergistic combinations are found to score above 28 while antagonistic scores fall below 8 consistently, and additive drug combinations score occupies a wide range of -8 to 28. 

The researchers have also taken the complex mechanism of combined drugs and chemical features of the high score of CSS into consideration as different combination therapies can have differential effects for each cancer type. The study has included melanoma, which is a type of childhood cancer and could be helpful in providing treatment recommendations. This approach has saved a lot of time by avoiding testing the drug combinations one by one, allowing healthcare professionals to divert their time to cases that require more attention.

3.3 Immunotherapy Side Effect Prediction Generated by Artificial Intelligence

Apart from the use of chemotherapeutic drugs, immunotherapy is also used commonly in paediatric cancer. Immunotherapy aids the immune system in fighting cancer by stimulating natural defences of the immune system, allowing it to work more efficiently to locate and attack cancer cells (American Cancer Society, n.d.). Immunotherapy is a type of biological therapy that uses substances derived from living organisms to treat cancer (National Cancer Institute, 2019). Immunotherapy is often used alongside other cancer treatments such as chemotherapy. 

AI algorithms can analyse a wide range of data, including genetic information and medical history, which includes previous health records and pathology reports, and scans like CT and MRI. These algorithms can identify the intricate pattern and correlation between data that holds the key to predicting the effectiveness of immunotherapy. Research has revealed the potential power AI may have in patients; this article (Hoque, 2023) has demonstrated how powerful AI is in patient care. AI can analyse a patient’s genetic profile and pinpoint mutations that make cancer vulnerable to immunotherapy. By spotting these critical biomarkers, healthcare workers are equipped with detailed data to make targeted decisions. 

Checkpoint inhibition is a type of immunotherapy. Checkpoints are the regulatory mechanism of the immune system that maintains the equilibrium of immune activity, preventing it from being too overactive and strong. By inhibiting these checkpoints, the immune system will work more vigorously to respond to cancer (National Cancer Institute, 2019). A model named LORIS (logistic regression-based immunotherapy-response score) was developed with the capacity to predict the patient’s likelihood of responding to immune checkpoint inhibitors based on patient information (age, cancer type, blood albumin and blood NLR) and medical history (history of cancer therapy) and estimate the short-term and long-term survival prospects after the immunotherapy is implemented (National Institutes of Health, 2024). This enhances the clinician’s ability to make personalised drug treatment. This is also particularly advantageous in paediatric cancer, where checkpoint inhibition therapy is given intravenously (Memorial Sloan Kettering Cancer Center, 2024). Young patients are more sensitive to discomforts from IV and may be distressed if they are unresponsive to the inhibitor. Clinicians can explore the most suitable immune checkpoint inhibitor, thereby optimising patient care and minimising unnecessary pain and distress. However, although these technologies can significantly alter the way cancer medication is created and personalised for patients, the healthcare system faces several challenges with the introduction of artificial intelligence into paediatric cancer care.

3.4 Paediatric Cancer Surgery and the Emergence of Artificial Intelligence

Surgery is another common approach for cancer in children involving solid tumours like Wilms tumour and neuroblastoma. Paediatric surgeons aim to remove as much of the tumour as possible while minimising damage to nearby unaffected healthy tissues. However, paediatric surgical oncology is challenging due to the location of children’s tumours and the high-risk procedure involved (Pio, L. and Sarnacki, S., 2022). The emergence of artificial intelligence may assist surgeons in these complex surgeries. 

In a hospital, a surgical robot Da Vinci Xi® was incorporated into paediatrics, with 149 cases in the oncology field (Vinit et al., 2024). In terms of renal tumours, the surgical robot has led to a smaller median tumour volume, shorter hospital stay, and reduced operative times compared to the traditional open surgical approach. This indicates that incorporating advanced technology can benefit the patient economically in addition to improved surgical outcomes. 

Furthermore, a 3D model was developed for the pelvis by using AI to analyse MRI images and provide a visual representation of the peripheral nervous system; it was then introduced into the hospital. Residents of the hospital reported that they find the model helpful in providing a better understanding of surgical steps planned, which would also assist their seniors in teaching and enhance teaching efficiency.

The use of artificial intelligence in treatment improves the overall patient experience by ensuring the effectiveness of the medicine used and minimising the undesired side effects. However, as artificial intelligence becomes more widespread in healthcare, concerns about patient privacy arise. Parents of the young patient are hesitant to have their sensitive medical records processed by AI systems. It may still take a long time before the use of artificial intelligence can be widely promoted within clinical practices. Though challenges remain, the healthcare industry is moving towards this direction.

4. Ethical issues Associated with the Use of AI in Paediatric Cancer

4.1 Data Security in Paediatric Cancer Care with AI

Artificial intelligence is a promising development and offers improved accuracy and can even be used to improve the efficiency of cancer medication combinations. However, there are significant ethical issues associated with the use of artificial intelligence in order to diagnose and treat cancer, especially in paediatric patients. Artificial intelligence trains using data sets of medical imaging to learn to distinguish between normal and abnormal scans and imaging. Over time, the engine can identify these abnormalities more accurately and eventually, the engine can then be used in real situations by a radiologist to help identify cancer where the results are unclear (Conner, 2024). 

However, the use of these datasets generates several issues. One challenge that comes with the use of artificial intelligence in paediatric oncology is the rarity of childhood cancer such as leukaemia, with paediatric cancer accounting for under 1% of cancer cases in the United States (eClinicalMedicine, 2024). This means that there are particularly small sample sizes and datasets to train the artificial intelligence engine when dealing with paediatric cancer, with around 60% of studies on paediatric cancer studying less than 100 patients and often using data from one individual for a study (Ramesh et al., 2021). 

This issue is only exacerbated by the ethical issue of data security. The use of datasets by artificial intelligence, although necessary, can be particularly dangerous as patient data must stay confidential between the doctor and patient. When dealing with underage patients, it is even more important to keep this information private as children are a dependent and vulnerable population and may become unsafe if their data is shared or breached, which is a risk when utilising artificial intelligence (eClinicalMedicine, 2024). Parents may feel uncomfortable with the knowledge that their children’s data can be accessed by artificial intelligence and therefore may not consent, meaning this dataset is reduced even further and cannot be trained, leading to inaccuracies and false results (Sisk et al., 2020). 

Finally, if the artificial intelligence engine used is trained on an adult demographic, due to anatomical differences between adults and children, the artificial intelligence may make certain assumptions which could lead to incorrect results (Hadhazy, 2023).

4.2 Informed Consent to AI in Paediatric Oncology

Another challenge that comes with the use of artificial intelligence in paediatric oncology is the issue of informed consent. With the development of artificial intelligence recently, informed consent has become a very important aspect of using these engines in cancer screening, with around 81% of oncologists believing that consent is necessary when using AI (Hantel et al., 2024). However, even the information and details offered in order to gather consent can be affected by artificial intelligence, as AI has the issue of ‘hallucinations’ and producing false results, meaning doctors may unintentionally provide patients with misinformation when asking for consent (Hryciw, 2023).

This issue surrounding informed consent is especially significant in paediatric oncology, as consent for a child’s procedure must be provided by a parent or guardian if the child is under 16 (National Healthcare Service, 2022). However, this can be very challenging to obtain from a guardian as it might be difficult to completely inform the parents about the use of artificial intelligence in their child’s cancer care, and due to the previously mentioned lack of datasets used for AI, many parents may not consent to an experimental system with the risk of inaccuracy (eClinicalMedicine, 2024). However, it is crucial that consent from both parents and patients is given with a clear understanding of the procedure. A study found that 91% of respondents to a questionnaire believe detailed information about the artificial intelligence process is necessary for parents to provide informed consent and 84% of these people would prefer these details to come from their doctor (Bergeah et al., 2024). 

Overall, although it is difficult to acquire informed consent and offer information about the procedure, it is vital that in paediatric cases, the patients and guardians understand the process, benefits and risks of using AI before consenting to an experimental method such as the use of artificial intelligence for diagnosis and treatment.

4.3 Trust and Doctor-Patient Relationship with the Use of AI in Paediatric Cancer Care

While parental guardians see treatment and accuracy as important for paediatric cancer care, they also see the relationship between patient and doctor as immensely significant and often place full trust in the doctor’s capabilities, especially with the challenges that come with a cancer diagnosis in a child. Almost 80% of parents claimed to absolutely trust their doctor after four months of their child’s cancer treatment, mostly generated by not only knowledge about their child’s treatment but also patient-centred treatment and interaction between the patient and doctor (Mack et al., 2020). However, the introduction of artificial intelligence into not only the analysis of medical imaging but also treatment plans as previously discussed has diminished this relationship and trust – a crucial aspect of cancer care, particularly in children. Many parents and guardians not only worry that artificial intelligence is inaccurate and could lead to the breach of data, but also that it leads to medical professionals relying on electronic records, computers and AI, and could therefore lead to reduced interaction between the doctor and patient. This could diminish genuine human engagement and therefore trust in the doctor’s capabilities (Sisk, 2020). Therefore, it is important that instead of solely relying on artificial intelligence for both analysis of medical imaging, treatment plans and patient care, doctors should use it sparingly, with patient care always remaining personal and maintaining a close and comforting relationship between the patient and doctor while also utilising the accuracy and objectivity of artificial intelligence, as artificial intelligence could still reduce the workload of doctors so they can spend time engaging with the patient and sustain this trust (Nagy, 2020).

4.4 Current Paediatric AI Guidelines and Possible Improvements

With the swift development of artificial intelligence in paediatric oncology alongside the risks and dangers of utilising this technology, it is important that guidelines and policies are provided to aid the use of artificial intelligence in paediatric healthcare. Currently, there are internationally-set guidelines for the use of artificial intelligence in both medical care and clinical trials known as the SPIRIT-AI and CONSORT-AI guidelines. These policies include guidelines on disclosing the procedure of acquiring patient data for artificial intelligence and reporting the demographic and possible sources of bias for the algorithm of the system (Ibrahim, 2021). Although these policies provide important guidelines for the use of this technology, they do not focus on the ethical issues surrounding the use of artificial intelligence in paediatric healthcare. In cases such as these, there has been a recent development of guidelines known as ACCEPT-AI. This new set of policies specifically focuses on the implications of using artificial intelligence in paediatric care and dealing with ethical issues that arise from treating those under 16 (Muralidharan, 2023). The first major aspect ACCEPT-AI focuses on is the lack of representation children have in clinical research and therefore the lack of a demographic to train the artificial intelligence with, leading to algorithmic bias as discussed in section 4.1. ACCEPT-AI details the solution to this, which is specific age-reporting and viewing the medical imaging and data of children as distinct to adults due to anatomical differences in development and structure (Muralidharan, 2023). Another significant aspect of ACCEPT-AI’s guidelines is concentrated on the topic of informed consent, which states that medical professionals should clearly communicate the process, risks, benefits, and alternative methods to artificial intelligence when gathering consent from parental guardians, while also disclosing the intentions of treatment plans and research. Furthermore, the guidelines include policies on data security, stating that first of all, data collected from a paediatric patient to either train or be analysed by the artificial intelligence engine must be strictly protected, and the legal processes involved must be very clear to the parental guardian and paediatric patient. Finally, these ACCEPT-AI guidelines mandate that the artificial intelligence technology must be constantly monitored and run through a cycle to assess the algorithm, any biases that could be present, and confirm that the artificial intelligence is functional (Muralidharan, 2023). Artificial intelligence has the ability to alter the way we diagnose and treat paediatric cancer and in order to take full advantage of this type of technology, these guidelines are very important and must become recognised as essential when this technology is eventually implemented into hospitals and the healthcare system.

5. Conclusion

In conclusion, artificial intelligence has a promising future in oncology. Artificial intelligence offers detail, objectivity, and accuracy to both diagnosis and treatment of paediatric cancer, allowing for a reduction of workload on the radiologist and doctor’s part, and further clinician-patient engagement. Furthermore, artificial intelligence has the capability to possibly shape the development of paediatric cancer treatment, a significant issue in cancer research. However, in order to use this technology, especially in a paediatric context, there must be guidelines detailing the procedure of using and protecting the patient’s data, ensuring that the data used does not result in algorithmic bias against the paediatric demographic. Without this, inaccuracies and false results may occur. Finally, when gathering consent, all parties involved should be fully informed of the details of utilising the technology and the process. If the guidelines surrounding the use of artificial intelligence can be agreed upon and properly adhered to, then this technology can enhance the precision and personalised nature of paediatric oncology care while keeping the patient safe and content.

Altomara, D. (2024) Chemotherapy: How it works and how you’ll feel. Available at: https://www.webmd.com/cancer/chemotherapy-what-to-expect (Accessed 20th August 2024)

American Cancer Society (2019) Immunotherapy. Available at: https://www.cancer.org/cancer/managing-cancer/treatment-types/immunotherapy.html (Accessed 20th August, 2024)

American Cancer Society (2024) Treating Children with Cancer. Available at: https://www.cancer.org/cancer/types/cancer-in-children/how-are-childhood-cancers-treated.html (Accessed 19th August 2024)

Berghea, E.C., Ionescu, M.D., Gheorghiu, R.M., Tincu, I.F., Cobilinschi, C.O., Craiu, M., Bălgrădean, M., Berghea, F. (2024) Integrating Artificial Intelligence in Pediatric Healthcare: Parental Perceptions and Ethical Implications. Children (Basel), 11(2). Available at: https://doi.org/10.3390%2Fchildren11020240 (Accessed 19th August 2024)

Cancer Australia. How cancer is diagnosed: Blood tests. https://childrenscancer.canceraustralia.gov.au/about-childrens-cancer/how-cancer-diagnosed/blood-tests (Accessed 22nd August 2024)

Cancer Australia. Medical imaging. https://childrenscancer.canceraustralia.gov.au/about-childrens-cancer/how-cancer-diagnosed/medical-imaging (Accessed 19th August 2024)

Children with cancer UK (n.d.) Chemotherapy. Available at: https://www.childrenwithcancer.org.uk/childhood-cancer-info/understanding-cancer/treatments/chemotherapy/ (Accessed 19th August)

Cleveland Clinic (2023) Childhood Cancer (Pediatric Cancer). Available at: https://my.clevelandclinic.org/health/diseases/24960-childhood-cancer (Accessed 18th August, 2024)

Conner, K. (2024) Using AI to Detect Breast Cancer: What We Know. BreastCancer.org. Available at: https://www.breastcancer.org/screening-testing/artificial-intelligence (Accessed 18th August 2024)

Dana-Farber Cancer Institue. (2018) What is the Difference Between Adult and Cancer in Children? https://blog.dana-farber.org/insight/2018/09/difference-cancer-adults-cancer-children/ (Accessed 19th August 2024)

Darlington, K. (2020) Factors associated with the detection of childhood and adolescent cancer in primary health care: A prospective cross-sectional study. Journal of Multidisciplinary Healthcare https://www.tandfonline.com/doi/full/10.2147/JMDH.S225641 (Accessed 23rd August 2024)

De Almeida, S.R., de Lima, C.M., de Oliveira, M.M., 2022. Timeliness of diagnosis and treatment: the challenge of childhood cancers. Rambam Maimonides Medical Journal. https://scielo.br/j/ramb/a/CpCzrKqgBQbZfCQyS63q9qg/?format=pdf&lang=en (Accessed 21st August 2024)

eClinicalMedicine (2024) The potential role of Artificial Intelligence in transforming childhood cancer care. The Lancet Discovery Science, 72. Available at: https://doi.org/10.1016/j.eclinm.2024.102705 (Accessed 18th August 2024)

El-Hafeez, T.A., Shams, M.Y., Elshaier, YA.M.M., Farghaly, H.M. and Hassanien, A.E. (2024) Harnessing machine learning to find synergistic combinations for FDA-approved cancer drugs. Available at: https://www.nature.com/articles/s41598-024-52814-w (Accessed 18th August, 2024)

Eldridge, L.(2024) An Overview of Myelosuppression, Consequences of Chemotherapy Induced Myelosuppression. Available at: https://www.verywellhealth.com/what-is-myelosuppression-224913 (Accessed 31 August, 2024)

García-Fuente, A., Vázquez, F., Viétez, J.M., Alonso, F.J.G., Martín, J.I.and Ferrer, J. (2018) CISNE: An accurate description of dose-effect and synergism in combination therapies. Scientific Reports 8, 4964(2018). Available at: https://doi.org/10.1038/s41598-018-23321-6 (Accessed 23rd August 2024)

Gurudu, S. D., Hwang, J. H., & Vezina, G. (2024) ‘Artificial Intelligence in Pediatric Cancer: Current Applications and Future Directions’, American Journal of Roentgenology, 223 (2), pp. 150-158. doi: 10.2214/AJR.24.31076. (Accessed 21st August 2024)

Hadhazy, A. (2023) Stanford Ethicists Developing Guidelines for the Safe Inclusion of Pediatric Data in AI-Driven Medical Research. Stanford University Human-Centred Artificial Intelligence. Available at: https://hai.stanford.edu/news/stanford-ethicists-developing-guidelines-safe-inclusion-pediatric-data-ai-driven-medical (Accessed 21st August 2024)

Hantel, A., Walsh, T.P., Marron, J.M., et al. (2024) Perspectives of Oncologists on the Ethical Implications of Using Artificial Intelligence for Cancer Care. JAMA Network Open, 7(3). Available at: https://doi.org/10.1001/jamanetworkopen.2024.4077 (Accessed 19th August 2024)

Hardie, R., Wang, L., Li, J., Sun, J., & Li, Z., 2023. Performance analysis in children of traditional and deep learning CT lung nodule computer-aided detection systems trained on adults. National Library of Medicine, 13(1), pp.1-11. Available at: https://pubmed.ncbi.nlm.nih.gov/37991333/ (Accessed 28th August 2024)

Houghton, J., 2021. Timeliness of Diagnosis and Treatment: The Challenge of Childhood Cancers. PubMed Central. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8651632/ (Accessed 23rd August 2024)

Hoque, F. (2023) AI for cancer treatment: better outcomes and lower costs. Available at: https://www.imd.org/ibyimd/technology/ai-for-cancer-treatment-better-outcomes-and-lower-costs/

Hryciw, B.N., Fortin, Z., Ghossein, J., Kyeremanteng, K. (2023) Doctor-patient interactions in the age of AI: navigating innovation and expertise. Front Med (Lausanne). Available at: https://doi.org/10.3389%2Ffmed.2023.1241508 (Accessed 20th August 2024)

Huang, Y., Yang, Y., Zhang, L., Zhang, Y., & Li, W., 2022. Artificial intelligence applications in paediatric brain tumour imaging: A systematic review. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/34648981 (Accessed 28th August 2024)

Ibrahim, H., Liu, X., Rivera, S.C., Moher, D., Chan, A.W., Sydes, M.R., Calvert, M.J., Denniston, A.K. (2021) Reporting guidelines for clinical trials of artificial intelligence interventions: the SPIRIT-AI and CONSORT-AI guidelines. Trials, 22(11). Available at:  https://doi.org/10.1186/s13063-020-04951-6 (Accessed 21st August 2024)

Król, M., Pawłowski, K.M., Majchrzak, K., Szyszko, K. and Motyl, T. (2010) Why chemotherapy can fail? Pol J Vet Sci, 13(2). Pp: 399-406. Available at: https://pubmed.ncbi.nlm.nih.gov/20731201/ (Accessed 18th August 2024)

Kuo, R.C., 2023. Artificial Intelligence Applications in Pediatric Oncology Diagnosis. PubMed Central. [online] Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10017189/ (Accessed 23rd August 2024)

Li, M., Wang, Q., Lu, P., Zhang, D., Hua, Y., Liu, F., Lin, T., Wei, G. and He, D. (2023) Development of a Machine Learning-Based Prediction Model for Chemotherapy-Induced Myelosuppression in Children with Wilms’ Tumour. Cancers (Basel), 15(4). Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9954251/  (Accessed 18th August 2024)

Li, X., Xu, Y., Cui, H., Huang, T., Wang, D., Lian, B., Li, W., Qin, G. Chen, L., Xie, L. (2017) Prediction of synergistic anti-cancer drug combinations based on drug target network and drug induced gene expression profiles. Available at: 10.1016/j.artmed.2017.05.008 (Accessed 18th August 2024)

Ma, L., Zhang, Y., Zhang, Z., & Chen, X., 2022. Machine learning-based CT radiomics method for identifying the stage of Wilms tumor in children. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/35676904/ (Accessed 28th August 2024)

Mack, J.W., Kang, T.I. (2020) Care experiences that foster trust between parents and physicians of children with cancer. Pediatric Blood Cancer, 67(11). Available at: https://doi.org/10.1002/pbc.28399 (Accessed 20th August 2024)

Martin, M. (2024) AI Harnesses Tumour Genetics to Predict Treatment Response. UC San Diego Today. Available at: https://today.ucsd.edu/story/ai-harnesses-tumor-genetics-to-predict-treatment-response (Accessed 18th August, 2024)

Memorial Sloan Kettering Cancer Center. (n.d.) Immune Checkpoint Inhibitor Cancer Treatment. Available at: https://www.mskcc.org/cancer-care/diagnosis-treatment/cancer-treatments/immunotherapy/checkpoint-inhibitors (Accessed 18th August, 2024)

Mirnezami, R. (2020) Cancer diagnostics and treatment decisions using artificial intelligence, In: Bohr, A. ed. and Memarzadeh, K. ed. Artificial Intelligence in Healthcare. Cambridge, Massachusetts. Academic press. Pp. 117-141. Available at: https://www.sciencedirect.com/science/article/pii/B9780128184387000058#s0040 (Accessed 19th August 2024)

Muralidharan, V., Burgart, A., Daneshjou, R. (2023) Recommendations for the use of paediatric data in artificial intelligence and machine learning ACCEPT-AI. npj Digital Medicine, 6(166). Available at: https://doi.org/10.1038/s41746-023-00898-5 (Accessed 21st August 2024)

Nagy, M., Sisk, B. (2020) How Will Artificial Intelligence Affect Patient-Clinician Relationships? AMA Journal of Ethics, 22(5), pp395-400. Available at: https://doi.org/10.1001/amajethics.2020.395 (Accessed 20th August 2024)

National Cancer Institute. (2024) Artificial intelligence and cancer. Available at: https://www.cancer.gov/research/infrastructure/artificial-intelligence (Accessed 18th August 2024)

National Cancer Institute (2024) Cancer in Children and Adolescents. National Cancer Institute. Available at: https://www.cancer.gov/types/childhood-cancers/child-adolescent-cancers-fact-sheet# (Accessed 31st August 2024)

National Cancer Institute (2019) Immunotherapy to treat cancer. Available at: https://www.cancer.gov/about-cancer/treatment/types/immunotherapy (Accessed 22nd August 2024)

National Cancer Institute, 2023. What Is Cancer?. Available at: https://www.cancer.gov/about-cancer/understanding/what-is-cancer  (Accessed 20 August 2024)

National Healthcare Service (2022) Children and Young People: Consent to Treatment. National Healthcare Service. Available at: https://www.nhs.uk/conditions/consent-to-treatment/children/ (Accessed 19th August 2024)

National Institutes of health. (2024) AI tool predicts response to cancer therapy. Available at: https://www.nih.gov/news-events/nih-research-matters/ai-tool-predicts-response-cancer-tLOherapy (Accessed 18th August 2024)

Ni, X., Xu, J., Lu, X., Zhang, Y., Zhang, Z., & Li, Q., 2023. Deep convolutional neural network based on CT images of pulmonary nodules in the lungs of adolescent and young adult patients with osteosarcoma. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/37427350/ (Accessed 28th August 2024)

Northeastern University College of Engineering. (2024) Using Machine Learning to Predict Cancer Treatment Response. Available at: https://coe.northeastern.edu/news/using-machine-learning-to-predict-cancer-treatment-response/ (Accessed 19th August 2024)

Pierra, S., Schaeffer, A., Coudray, N., Cornu, P., Braccini, S., & Robert, J., 2021. Machine learning-based CT radiomics features for the prediction of pulmonary metastasis in osteosarcoma. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/34111978/. (Accessed 28th August 2024)

Pio, L. and Sarnacki, S. (2022) Editorial: Innovative approaches in paediatric surgical oncology. Front Pediatr, 10(989822). Available at: 10.3389/fped.2022.989822

Ramesh, S., Chokkara, S., Shen, T., Major, A., Volchenbou, S.L., Mayanpurath, A., Applebaum, M.A. (2021) Applications of Artificial Intelligence in Pediatric Oncology: A Systematic Review. JCO Clinical Cancer Informatics, 5, pp1208-1219. Available at: https://ascopubs.org/doi/10.1200/CCI.21.00102 (Accessed 18th August 2024)

Scott, J. (2023) Medical News Today. Survival and chemotherapy success rates for various cancers. https://www.medicalnewstoday.com/articles/326031#success-rates-overview (Accessed 31st August 2024)

Siegel, D.A., Richardson, L.C., Henley, J.S., Wilson, R.J., Dowling, N.F., Weir, H.K., Tai, E.W. and Lunsford, N.B. (2020) Pediatric Cancer Mortality and Survival in the United States, 2001–2016. Cancer, 126(19) pp. 4379-4389. Available at: 10.1002/cncr.33080 (Accessed 23th August 2024)

Singleman, C. (2024) Ai Screening of Pediatric Cancers Improves Treatment outcomes. Genetic Engineering & Biotechnology News. Available at: https://www.genengnews.com/topics/cancer/ai-screening-of-pediatric-cancers-improves-treatment-outcomes/ (Accessed 19th August 2024)

Sisk, B.A., Antes, A.L., Burrous, S., DuBois, J.M. (2020) Parental Attitudes toward Artificial Intelligence-Driven Precision Medicine Technologies in Pediatric Healthcare. Children, 7(9), pp145. Available at: https://doi.org/10.3390/children7090145 (Accessed 18th August 2024)

Teo, M.T., Bown, N., Martin, J.A., Evans, A., Firth, A., & Smith, M.A., 2022. Correlation of histopathology and multi-modal magnetic resonance imaging in childhood osteosarcoma: Predicting tumor response to chemotherapy. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/35157711/ (Accessed 28th August 2024)

UK Health Security Agency (UKHSA), 2017. Childhood cancer statistics: What can we learn from new data? https://ukhsa.blog.gov.uk/2017/02/01/childhood-cancer-statistics-what-can-we-learn-from-new-data/ (Accessed 20th August 2024)

U.S. Environmental Protection Agency (EPA). (n.d.) Health and childhood cancer. https://www.epa.gov/americaschildrenenvironment/health-childhood-cancer (Accessed 23rd August 2024).

Vinit, N., Blanc, T., Bloch, I., Pio, L., Kassir, R., Barbera, G.L., Bonnot, E., Gori, P., Goulin, J., Pire, A., Boddaert, N., Lozach, C. and Sarnacki, S. (2024) Robotics and 3D modelling for precision surgery in paediatric oncology. EJC Paediatric Oncology, 4(100181). Available at: https://doi.org/10.1016/j.ejcped.2024.100181

Wang, L., Zhang, J., Liu, X., Yang, Y., & Wang, Y., 2023. Multimodal paediatric lymphoma detection using PET and MRI. National Library of Medicine. https://pubmed.ncbi.nlm.nih.gov/38222333/. (Accessed 28th August 2024)

Yale Medicine, 2024. Paediatric cancer treatment. https://www.yalemedicine.org/conditions/pediatric-cancer-treatment (Accessed 21st August 2024)

Yang, Y., Liu, S., Chen, Y., Wang, X., & Zhang, L., 2023. Artificial intelligence applications in paediatric oncology diagnosis. National Library of Medicine https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10017189/ (Accessed 28th August 2024)

Zheng, K. (2023) Artificial intelligence and cancer care: the future of oncology? Stanford medicine. Available at: https://med.stanford.edu/cancer/about/news/ai.html (Accessed 18th August 2024)