Abstract

Ovarian cancer is one of the most lethal types of cancer and is the leading cause of death in women diagnosed with gynecological cancers (Arora et al., 2024). This paper delves into the complex nature of ovarian cancer, exploring its pathophysiology and the mechanisms that drive its metastasis, as well as the broader impact the disease has on the body. The study provides a comprehensive understanding of the disease’s origins and progression by examining common risk factors. Additionally, the paper investigates the role of artificial intelligence (AI) in managing ovarian cancer, particularly in enhancing diagnostic accuracy through advanced techniques such as ultrasounds, liquid biopsies, and imaging modalities like CT and MRI scans. Additionally, the use of AI in the treatment process is analysed, specifically in predicting the efficacy of drug therapies, the mutation status of genes, and different pathological subtypes to inform the development of chemotherapy drugs and treatment plans. The paper concludes by discussing recent advancements and discoveries in AI that have promoted the detection and management of ovarian cancer, which can significantly improve mortality rates and outcomes of cancer patients.

Introduction

Affecting 1 in 87 women, ovarian cancer accounts for 2.5% of cancers in all women and is the fifth most common cause of death from cancer (Ovarian Cancer Research Alliance, 2024). Approximately 324,000 women are diagnosed with ovarian cancer, and 207,000 women die from it every year (World Ovarian Cancer Coalition, 2024). The stage at diagnosis plays an integral part in the five-year survival rate. At Stage 1, the survival rate is around 95%; at Stage 2, it decreases to 70%; it takes a significant dip, making it 25% at Stage 3 and 15% at Stage 4 (Cancer Research UK, 2024). Overall, it is responsible for 2.1% of all cancer deaths, though the mortality rate has decreased by 40% since 1975. Ovarian cancer is the deadliest type of gynecologic cancer, only affects biological women, and develops rarely in women before the age of 40; over half of ovarian cancer patients are aged 63 or older (American Cancer Society, 2024). 

Artificial intelligence (AI) has been used to diagnose and treat patients with cancer in recent years. In breast cancer detection, for example, physicians use the help of AI to analyse mammograms as well as biopsies, making the diagnosis more efficient and less time-consuming (Jiang et al., 2024). AI is also used in cancer treatment; in prostate cancer treatment, for example, researchers at Oxford University were able to develop a deep learning machine that created personalised treatment schedules and strategies for treatment, while also being robust to the patient’s response to the treatment and time-interval of treatments (Gallagher et al., 2024). In this manner, AI has been developed to become a part of the medical field. However, there has been debate on the ethics of using AI in the medical field, making it challenging to implement it to become a part of the clinical workflow. 

The field of AI has not yet been adapted to ovarian cancer. There is little research available on these algorithms, and many algorithms in certain scans and certain types of treatment are not yet available to ovarian cancer patients. This review aims to cover the role of AI in the diagnosis and treatment of ovarian cancer and reviews some significant recent developments in the field, as well as necessary ethical considerations to be made to implement this technology into the medical field.

Background on Ovarian Cancer

Ovarian cancer is a type of cancer characterised by the uncontrolled growth of harmful cells in the ovaries. Ovarian cancer can be divided into subtypes, depending on whether the cancer has an epithelial or non-epithelial origin. An epithelial origin denotes that the cancer started in the epithelium of the ovaries or the thin layer of body tissue covering the surface of the ovaries. In non-epithelial origin, the cancer is started from the egg cells in the ovaries. This subtype is rarer; only 10% of all ovarian cancers fall under this category, while 90% are epithelial (Momenimovahed et al., 2019). The epithelial subtype is also categorised into mucinous and non-mucinous cancers. Mucinous cancer is sporadic, accounting for 3% of epithelial cancers, and it is when the tumour cells are covered in mucus, which protects them from the immune system when the tumour metastasizes or travels to other parts of the body. Non-mucinous cancers can also be further classified into serous (tumour originates from the serous membrane, which is the outer lining of organs and body cavities of the abdomen and chest), endometrioid (tumour forms in the endometrium, or the lining of the uterus), and several other subtypes (Momenimovahed et al., 2019).

The spread or the metastasis of ovarian cancer typically follows the lymphatic vessels in the body, most commonly reaching the para-aortic and paracaval lymph nodes first and then the external iliac, common iliac, hypogastric, and lateral sacral lymph nodes (Arora et al., 2024). The most common distant metastasis site is the peritoneum, a thin tissue covering abdominal organs. When ovarian cancer spreads to the fallopian tubes, the cancer can grow into a tubal intraepithelial carcinoma. This spread may appear as a widespread malignancy involving both the ovary and the fallopian tube, and it can be difficult to determine whether the primary tumour started in the ovary, fallopian tube, or peritoneum. The exact causes of ovarian cancer are not fully understood, but genetic mutations play a significant role. These mutations can occur either before birth (germline) or after birth (somatic). A common somatic mutation in ovarian cancer is in the TP53 gene, especially in high-grade serous carcinomas. Other less common somatic mutations include CSMD3, FAT3, BRCA1, BRCA2, PTEN, PIK3CA, KRAS, BRAF, CTNNB1, and PPP2R1A (Arora et al., 2024). Epigenetic changes, such as DNA methylation, which affects how genes are regulated, are also linked to the development of ovarian cancer and other cancers. Lynch syndrome, also known as hereditary nonpolyposis colorectal cancer, is associated with mutations in mismatch repair genes (MLH1, MSH2, MSH6, PMS2, and EPCAM) that increase the risk of ovarian cancer. Other inherited conditions related to ovarian cancer include Cowden syndrome (PTEN mutation) and Peutz-Jeghers syndrome (STK11 mutation) (Arora et al., 2024).

There are several risk factors related to ovarian cancer, one of which is age, as it is more common in women over 65 years (Momenimovahed et al., 2019). Some racial groups are more affected by it than others, and the highest prevalence of ovarian cancer is seen in non-Hispanic White women (12.0 per 100,000), followed by Hispanic (10.3 per 100,000), non-Hispanic Black (9.4 per 100,000), and Asian/Pacific Islander women (9.2 per 100,000) (Momenimovahed et al., 2019). The incessant ovulation theory states that ovulation without interruption can increase the chance of ovarian cancer, as it damages the epithelium of the ovaries. Accordingly, a study conducted has shown that any factor that reduces ovulation, including pregnancy, can protect against ovarian cancer (Momenimovahed et al., 2019). Other characteristics of pregnancy can also influence the chances of developing cancer, such as older age pregnancies and a large number of live births are associated with decreased risks, and preterm labor can increase the risk. Additionally, lactation and the duration of breastfeeding are also correlated with lower possibilities of ovarian cancer. There are many genetic risk factors, such as the personal history of breast cancer, mutations in the BRCA genes, and inheriting Lynch syndrome (Momenimovahed et al., 2019). Several lifestyle habits like smoking, obesity, and consumption of certain types of alcohol have been known to severely increase the probability of developing ovarian cancer as well (Momenimovahed et al., 2019).

Background on AI Use in Medicine

Artificial intelligence is fundamentally a technology that can perform tasks that require human intelligence. AI systems used in the medical field usually follow a pattern – such a system starts with a large amount of data, and on that data, machine learning algorithms are employed to gain information; this information is then used to generate a useful output to solve a well-defined problem in the medical system (Basu et al., 2020). AI is commonly used in healthcare to diagnose and prognose (predict disease progression) patients and match them with the right physicians. It is also used to translate languages, transcribe notes, and organise files. Applications of artificially intelligent systems in healthcare can be broadly classified into three categories: Patient-oriented AI (systems that directly assist and interact with patients), Clinician-oriented AI (systems that assist healthcare providers), and Administrative and Operational-oriented AI (systems for administration and operations like automated scheduling, billing, and others) (Basu et al., 2020). Ultimately, AI is expected to enhance the healthcare sector in various ways, including patient diagnosis, prognosis, and drug discovery, and it will act as an assistant to physicians, ultimately providing a more personalised and improved patient experience.

Use of AI for Diagnosis of Ovarian Cancer

i. Diagnosis

There are multiple ways of diagnosing ovarian cancer. Pelvic exams, ultrasounds, and CA125 (Cancer Antigen 125) tests are the main tests to diagnose the existence of the cancer, while CT scans, as well as MRI scans, are also used, mainly to understand the staging of the cancer. A physician would typically listen to the patient’s symptoms and perform a pelvic exam. This checks if there is an abnormal growth or enlarged organs, some common effects of ovarian cancer. The doctor could order additional tests like ultrasounds and CA125 tests. For ultrasounds, a pelvic or transvaginal ultrasound is conducted. They can perform other imaging tests such as MRIs and CT scans, as well as a PET scan. CA125 tests are often used to diagnose ovarian cancer as another method to evaluate the level of the cancer antigen in the bloodstream. However, CA125 in the bloodstream can be normal, even when cancer is present, so it is usually conducted alongside an imaging exam. Biopsies are another way ovarian cancer is diagnosed; it is most commonly done by surgically removing the tumour. Another uncommon test conducted is a genetic test to determine if the patient is a carrier of a certain gene that could put them at greater risk for ovarian cancer. This article mainly focuses on imaging and blood tests.

ii. Ultrasounds

Ultrasounds are challenging to read for AI because they are versatile compared to other screening types, such as CTs and MRIs (Beacher, n.d.). AI-based ultrasounds are highly dependent on operators scanning the patient. The operator must choose a specific organ or area of interest to scan, and each part has its data variability. Data subjectivity is another issue; the scans’ accuracy solely depends on the operator’s experience and skills. Furthermore, ultrasounds sometimes require different positioning and inspiration (Wieser, 2023). The limited quality of ultrasound images poses a challenge, especially if the abnormality is minor or deep inside the body (Tenajas et al., 2023). It would be challenging for AI to find data on each circumstance because of how versatile ultrasounds can be, and without the data, it cannot create an accurate analysis. 

However, AI for ultrasound was developed to adapt to the versatility of these scans. To create the most accurate scans possible, operators must place the transductor in a precise location on the patient’s body. One algorithm that solves this issue is the You Only Look Once (YOLO) algorithm (Coelho, 2023). YOLO places coordinates on the image, predicting the probability of an object’s presence and the image’s bounding box coordinates (Mirkhan, n.d.).

When the operator scans the patient’s body, the AI will recognise the scan and assess the image quality. When the ultrasound device creates a 2D image, AI assists with beam folding, higher resolution, and image enhancement. Then, AI is used to assist the operator in measuring and quantifying specific abnormalities and finally diagnose the patient (Tenajas et al., 2023).

iii. Liquid Biopsy

CA125 blood tests are the most common blood tests used in the diagnosis of ovarian cancer. However, they can be inaccurate and create false positives, especially when the patient is menstruating, is pregnant, has endometriosis, pelvic inflammatory disease, etc. (Cleveland Clinic, 2022). Because of its frequency of creating false positives, CA125 tests must be performed alongside another test, like an ultrasound. To resolve this inconvenience, a new liquid biopsy test was developed to alleviate this issue. Liquid biopsy is a tool to find cancerous DNA fragments in the blood. Liquid biopsies can detect circulating tumour cells (CTCs) and tumour DNA (cDNA). Liquid biopsies have lower sensitivity compared to tissue biopsies. However, using the DNA evaluation of fragments for early interception (DELFI) model, Johns Hopkins researchers combined the biopsy data with CA125 and human epididymis protein four. They put the results through an algorithm-trained machine and found that the algorithm had a 99% success rate – a significant level of accuracy compared to other methods (Cleveland Clinic, 2022). 

iv. CT Scans

It takes 30 to 40 minutes on average for CT scans to be read, and they can only be seen from one dimension at a time. These create false perceptions in the radiologist of static or minimal changes when it could be a more significant issue than that presented on the scans. Using AI, specifically Graphic User Interface (GUI) technology, radiologists can predict prognostic features when they receive scans from CT devices, and these scans can be analysed in 2D or 3D, making it easier to compare actual lesion size while also creating calculations of the measurements (Roke n.d.). The AI could section off pieces of different tissue using the texton classic computer vision technique. These contents are analysed based on the information inputted and AI algorithms through technology such as Facebook’s self-supervised method called “DINO”. Deep learning tools are also utilised to expand 2D CT scans to 3D scans, making it more straightforward for physicians to have a clear view of the lesions and create a treatment plan (Roke, n.d.). Furthermore, in a comparison between the reading of CT scans for lung cancer between AI and human radiologists, no significant difference was noted between AI and radiologists (Jacobs et al., 2021). These results show promise for further implementation of AI in the field, and ovarian cancer especially after further development and adjustment. 

v. MRI Scans

MRI scans are not recommended to create an initial diagnosis of ovarian cancer. However, they can be taken to assist in staging the condition and creating a better treatment plan for the patient. MRIs, however, take a long time to create and are very costly per scan. Meta developed the fast MRI, a raw collection of MR measurements and images, and trained it with NYU Langone Health’s database. It now creates scans based on parts of MRI scans with minimal time and accurate data (Zbontar, 2018). With Microsoft’s Apoqlar, AI can create virtual reality software programs, allowing surgeons a better view of the situation and a better plan for surgery. It blends MRI scans, 3D imaging, clinical workflow, and medical education. Furthermore, a deep learning model based on ​​convolutional neural networks was used to differentiate malignant and benign tumours. They performed equivalent to radiologists (Saida et al., 2022). MRI, like other tests, is becoming adapted to artificial intelligence, proving to be as accurate as human radiologists.

Use of AI for Treatment of Ovarian Cancer

i. Treatment

The treatment approach for ovarian masses can vary significantly, with the primary treatments being surgery (hysterectomy, unilateral salpingo-oophorectomy, laparotomy, etc.) and chemotherapy. Other options like targeted medications and hormone therapies may also be considered, depending on the preliminary diagnosis from imaging tests, even before a confirmative histological diagnosis is obtained (Dana-Farber, n.d.).

Patients with a family history of ovarian cancer, postmenopausal status, and solid components on scans may have increased risks of ovarian malignancy and require treatments in gynecology centres with appropriate staging and where radical surgery can be offered. 

The accurate pre-operative characterisation of ovarian tumours is therefore crucial to improving patients’ outcomes and reducing the morbidity burden of the disease.

ii. Prediction of Different Pathological Subtypes for Different Chemotherapy Drugs and Treatment Plans

A pathological biopsy during postoperative chemotherapy is essential for diagnosing epithelial ovarian cancer, which is the most common type of cancer that develops in the epithelial tissue outside of an ovary. The therapeutic approach for ovarian cancer varies based on its pathological subtypes, which require different chemotherapy regimens. Hence, clinicians can create more personalised treatment plans with the pathological findings and biopsy results (Cleveland Clinic, n.d).

The conventional method involves staining tissue samples with hematoxylin and eosin (H&E), and examining them under a microscope. Whole Slide Imaging (WSI) converts pathological tissue sections into high-resolution digital images. This method is subjective and reliant on the pathologist’s experience and includes the difficulty of storing tissue slices post-diagnosis and limitations in remote consultations. However, identifying subtypes often relies on pathologists’ subjective judgment, with low reported interobserver consistency, addressing the limitations of traditional methods and enhancing diagnostic efficiency and accuracy.

Deep learning has been increasingly utilised in medical pathological image recognition, improving the digitisation of pathology and aiding in the analysis of pathological images. For instance, the Convolutional Neural Network (CNN) algorithm, developed by Farahani et al. is used for Subtype Identification and achieved high concordance with pathologists in identifying OC subtypes (81.38% in training and 80.97% in external datasets).

Moreover, Wang et al. created a weakly supervised DL model that accurately predicts the efficacy of bevacizumab in OC treatment by analysing histological images and guiding clinical decisions for patient management. Integrating whole-slice models with deep learning can extract significant information from high-throughput pathological data, enhancing precision in treatment planning.

iii. Predict the Efficacy and Prognosis of Drug Therapy

Currently, the standard treatment for EOC is cytoreductive surgery combined with platinum-based chemotherapy. Cytoreductive surgery is a set of peritonectomy procedures and visceral resections for the complete removal of all visible diseases from the abdomen and pelvis. In the context of ovarian cancer, this surgery may include total hysterectomy (removal of the uterus), bilateral salpingo-oophorectomy (removal of both ovaries and fallopian tubes), omentectomy (removal of the omentum), and resection of any other visible cancerous tissues or organs as needed (Moffitt, n.d). Combined with platinum-based chemotherapy (drugs containing platinum ion compounds), this approach aims to remove all visible tumours and optimises the likelihood of long-term survival for patients by targeting any remaining cancer cells that may not have been removed during surgery (Royal United Hospital Bath, n.d.).

Patients with different pathological types of OC have different sensitivity levels to platinum-based chemotherapy, which are classified into platinum-sensitive or platinum-resistant according to the time from the end of treatment to the recurrence of the disease (platinum-free interval). For instance, ovarian cancer with a reoccurrence in six months after first-line platinum-based chemotherapy is considered platinum-resistant. According to Ghofraan Abdulsalam Atallah et al., the likelihood of platinum resistance occurs in approximately 25% of cases in individual patients diagnosed with ovarian cancer, and the median progression-free survival (PFS) is only 9-12 months on average (Ghofraan Abdulsalam Atallah et al., 2023).

The effectiveness of platinum-based chemotherapy has decreased due to the emergence of chemotherapy resistance and refractory diseases. However, treatments can be modified utilising AI algorithms to detect the individual patient’s sensitivity to platinum. The proposed deep learning model can differentiate patients likely to respond well to treatment from those with a higher risk of recurrence or disease deterioration (Yanli Wang et al., 2024).

The CNN model mentioned in the previous section uses whole slide imaging of high-grade serous ovarian cancer (HGSOC) patients who underwent platinum-based chemotherapy to predict treatment outcomes. The CNN model effectively distinguished between patients with different responses to platinum drugs, achieving a sensitivity of 73% and specificity of 91% (Laury et al., n.d.). 

The model allows for the analysis of survival rates based on different homologous repair deficiency states among patients, aiding in overall risk stratification. The deep learning model helps in risk stratification and distinguishes between different subtypes of OC, providing a basis for targeted therapy.

Moreover, in collaboration with the University of Cambridge, Professor Evie Sala’s team developed an AI-based model – IRON (Integrated Radiogenomics for Ovarian Neoadjuvant Therapy) – that further demonstrates the ability of AI to modify targeted treatment plans for ovarian cancer. This model can measure the volumetric reduction of tumour lesions in 80% of ovarian cancer patients. With an accuracy of 80%, the model analyses various patient clinical features to predict the therapy outcome – from liquid biopsy to general characteristics (age, health status, etc.), tumour markers, and disease images obtained through CT scans (Crispin-Ortuzar, M. et al., 2023).

New Advancements in Ovarian Cancer Diagnosis and Treatment

The integration of AI in ovarian cancer diagnosis and treatment is still a developing field. This section provides an overview of the emerging technologies anticipated to be seen in the field soon, assisting healthcare practitioners with diagnosis and treatment and benefitting the patients. Many new tests and machines help clinicians provide the best patient care, from diagnosis to treatment. 

i. Comprehensive Serum Glycopeptide Spectra Analysis Combined with Artificial Intelligence

Serum protein tumour markers, widely used, evaluate the measurement of molecules in the assay system based on antigen-antibody reactions to a single molecule. However, many issues, such as the developing state of tumour markers, pose difficulties in creating an accurate diagnosis, and the researchers at Japan’s LSI Medical Solutions created better alternative methods of diagnosing ovarian cancer. This exam aims to identify cancer by a single tumour marker or use a diagnostic system that identifies pathological conditions using all peak value data, creating the ultimate combination assay by collecting glycoproteins from blood tests and inserting them in Liquid Chromatography/Mass Spectrometry (LC/MS), which extracts approximately 2,000 reproducible peak values from the two-dimensional data obtained from the LC/MS, and compares that data between patients identified by a single tumour marker. Furthermore, the research group developed two new algorithms to diagnose early-stage ovarian cancer from just serum (Tanabe et al., 2020).

ii. Radiomics in Metastasis Detection

Early detection of metastases in ovarian cancer is crucial for effective treatment and management, but conventional methods often miss small metastases (Yu et al., 2024). Radiomics offers a new approach to improve metastasis detection. For example, Ai and colleagues identified specific radiomic features from CT images that, when combined with clinical factors like age and CA125 levels, effectively predicted metastatic status in ovarian cancer patients (Ai et al., 2024). Similarly, another study developed a nomogram using radiomics characteristics and clinical data from FS-T2WI, DWI, and dynamic CE-MRI images, which exceeded traditional clinical models in predicting peritoneal metastasis (Yu et al., 2024). These findings highlight the potential of radiomics to enhance the early detection of tiny metastatic lesions, which is crucial for timely intervention and can significantly improve overall patient outcomes by enabling more personalised and effective treatment strategies.

iii. Deep Learning Models in Ovarian Tumour Diagnosis

Currently, benign and malignant ovarian tumours are differentiated through the pathological analysis of a puncture biopsy or postoperative pathological examination. New approaches utilising radiomics to distinguish the two types of tumours have been developed, offering a non-invasive alternative to traditional pathological analysis.

Deep learning (DL) models have also been applied to ovarian cancer diagnosis, as they have been shown to surpass the diagnostic accuracy of radiologists. DL models can recognise OC subtypes at the molecular scale, and can pick out specific target genes that are associated with the molecular subtypes (using differential expression analysis and weighted gene co-expression network analysis). Saida et al. revealed that MRI-based deep learning models, such as CNNs (convolutional neural networks, machine learning models), provided better diagnostic efficiency than radiologists. DL models based on various imaging techniques, including CT and ultrasound, provide high sensitivity and specificity in distinguishing malignant from benign ovarian tumours, making them valuable tools in the clinical space.

Ethical Considerations of Artificial Intelligence in the Field of Ovarian Cancer

The following ethical considerations for using AI in ovarian cancer research are based on the World Health Organisation’s fundamental ethical principles outlined in “Ethics & Governance of Artificial Intelligence for Health”. These considerations are crucial for guiding future research to enhance the benefits of AI, while minimising risks and avoiding potential pitfalls. This section provides a framework for governments, technology developers, companies, civil society, and intergovernmental organisations to implement ethical practices in AI applications for ovarian cancer research (World Health Organisation, 2021).

i. Protect Autonomy

Autonomy is defined as acknowledging the right of patients with decision-making capacity to make informed decisions about their medical care, even in scenarios that contradict clinicians’ recommendations (British Medical Association, 2024).

When used in medical practices, AI algorithms inherit the possibility of transferring decision-making power to computational systems without the patient’s consent. Therefore, when creating future AI models for ovarian cancer detection and treatment, human practitioners must be involved in decision-making autonomy and oversee AI-driven healthcare practices to guarantee algorithms do not overturn patient decisions. 

Moreover, medical practitioners must protect the privacy and confidentiality of individual patients when diagnosing with AI. AI inherently requires large amounts of data, such as meticulously curated and annotated mammographic images, which can be used to train and test deep learning algorithms. With the current knowledge of AI, practitioners must acknowledge that the patients may be at risk of data with sensitive health information being exposed to the public in transit, at rest, or in use without the permission of the concerned patient. 

Thus, protecting patient information is essential to further research, such as establishing a standardised protocol for data encryption and sharing in AI-based research. Technicians are also generally advised to retrospectively use anonymised patient treatment data, such as delinking the meta-data of patients when using it for AI studies (Neel Yadav et al., 2023).

ii. Foster Responsibility and Accountability

Like human physicians, AI can miss tumours in scans or advise the wrong amount of drugs for treatment, putting the patient at risk and at a disadvantage. Hence, stakeholders are responsible for ensuring the algorithm has been monitored and trained under appropriate conditions having the capacity to perform specific tasks. 

While there have been no recorded cases of AI medical malpractice, research has been done from two perspectives: one that supported the implementation, and one that did not. A study conducted by Price et al. points out that physicians must understand the thought process of these AI algorithms, figure out how much of the recommendation from the AI they will take, and what type of medical AI they should use. However, this is still in the process of being developed. The Food and Drug Association (FDA) points out that conferences and organisations need to create additional guidelines concerning the patients’ needs and think about no-responsibility policies concerning AI in the future. Tobia et al. concluded that in AI treatment, whether that suggestion was a general textbook solution or a non-standard treatment option and whether the physician followed the guide from the AI were heavily considered (Funabashi, 2020).

Medical malpractice must be significantly considered, especially when human physicians are not entirely involved. Hence, human subversion is needed during the process, and related medical professionals are accountable for the AI technology if anything goes wrong. 

Furthermore, necessary guidelines on the use of AI must be created by the government, and appropriate mechanisms should be available for questioning individuals and groups that are adversely affected by decisions based on algorithms. The occasional inaccuracy of AI must be considered when implementing the technology into the clinical workflow.

iii. Ensure Inclusiveness and Equity

The WHO states that AI should support fair access for all people, regardless of age, gender, colour, or other traits. With different models/advancements of AI, it is undeniable that they have the power to transform healthcare. However, integrating AI with inclusivity and equity must be given top priority. 

The potential for algorithmic bias to permeate healthcare AI is significant and multifaceted – such as unintentionally favouring particular demographic groups by algorithms, which can result in unequal resource allocation and exacerbate already-existing imbalances. 

The AI algorithms are frequently trained on datasets that perpetuate biases that might erode public confidence in healthcare systems by reflecting societal injustices and resulting in a deficiency of diversity. Ziad Obermeyer et al. suggested that the algorithm projected healthcare expenses equitably but led to fewer referrals for Black patients than their White counterparts due to Black-sounding names being more likely to trigger ads for arrest records and with existing data having fewer features from Black women. Marginalised communities may grow sceptical about AI-driven services in the long run without this focus, resulting in reduced trust in patients and practitioners in general. 

While there is deficient research conducted about the AI models in ovarian cancer, this seems to be a common concern with AI in general. A case study by Vyas, et al., conducted on caesarean delivery (C-section), highlights the algorithmic bias of a Vaginal Birth After Caesarean Delivery (VBAC) that mispredicted the likelihood of safe birth through vaginal delivery, with a significantly higher population of Black/African-American and Hispanic/Latina women and causing doctors to perform more C-sections on Black/African-American and Hispanic/Latina women than on White women. When modified, the new version of the algorithm no longer considers race or ethnicity when predicting the risk of complications from VBAC. With the assistance of this new model, doctors can make decisions based on more accurate and impartial information that works for the individual patient, providing more equitable care regardless of race or ethnicity (Caleb J. Colón-Rodríguez, n.d.).

This case study further demonstrates the current limitations and the possible consequences of AI models with potential algorithmic bias and inequity that may make patients sceptical of AI in general – the same applies to AI models in the field of ovarian cancer. To mitigate the inequity within the healthcare system, which is exacerbated by the use of AI, federal governments are suggested to establish and implement standards to regulate algorithms and to assess external factors, such as race, ethnicity, gender, and more, that are directly influencing diagnosis, treatment, and access to care of patients (Federation of American Scientists, n.d.).

Conclusion

Overall, this article has discussed the complexity of ovarian cancer, highlighting the various genetic mutations that contribute to its different types and metastatic mechanisms of action. We also explored the critical risk factors that impact early detection and treatment. Furthermore, we examined the role of AI in revolutionising cancer care, mainly through advancements in medical imaging and diagnostics. Artificial intelligence is being introduced to the field of cancer diagnosis, with its ability to process large datasets and to be extensively utilised in developing diverse omic models for ovarian cancer. Multi-omics analysis, including imaging, pathomics, genomics, metabolomics, and proteomics, has demonstrated potential in enhancing the accuracy of OC diagnoses, differentiating benign and malignant cases, and predicting pathological types and prognosis. These innovations can result in more accurate and personalised approaches to ovarian cancer treatment, ultimately improving patient outcomes in the future. 

Despite being a prevalent cancer with global significance, further research may have to be conducted with government implementation, such as developing accredited deep learning artificial intelligence model policies devoted to this disease. The integration of multi-omics data has the potential to improve patient survival and facilitate precision medicine in the future.

Arora, T., Mullangi, S., Vadakekut, E.S. and Lekkala, M.R., 2024. Epithelial Ovarian Cancer. StatPearls.

American Cancer Society. (n.d). Key Statistics for Ovarian Cancer. Available at: https://www.cancer.org/cancer/types/ovarian-cancer/about/key-statistics.html. (Accessed 31 August 2024).

Advances in artificial intelligence for the diagnosis and treatment of ovarian cancer (Review). (2024). NCBI. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10828921/. (Accessed 1 September 2024).

Cancer Research UK. (n.d.). Ovarian Cancer Survival. Available at: https://www.cancerresearchuk.org/about-cancer/ovarian-cancer/survival. (Accessed 31 August 2024). 

Cleveland Clinic. (n.d.). CA-125 Blood Test. [online] Available at: https://my.clevelandclinic.org/health/diagnostics/17830-ca-125-blood-test. (Accessed 31 August 2024). 

Cleveland Clinic. (n.d.). Liquid Biopsy: What It Is & Procedure Details. [online] Available at: https://my.clevelandclinic.org/health/diagnostics/23992-liquid-biopsy. (Accessed 31 August 2024).

Felix, B. (n.d.). Information Classification: General. [online] Ultrasound and AI – Where are We? Available at: https://www.bmus.org/static/uploads/resources/Ultrasound_and_AI__Where_are_We.pdf. 

Funabashi, A. Jacobs, C. , Setio, A.A.A., Scholten, E.T., Gerke, P.T.,  Bhattacharya, H., Hoesein, F.A.M., Brink, M., Ranschaert, E., de Jong, P.A., Silva, M., Geurts, B., Chung, K., Schalekamp, S., Meersschaert, J. Deep Learning for Lung Cancer Detection on Screening CT Scans: Results of a Large-Scale Public Competition and an Observer Study with 11 Radiologists. Radiological Society of North America. doi:https://doi.org/10.1148/ryai.2021210027. 

Gallagher, K.,  Strobl, M.A.R., Park, D.S., Spoendlin, F.C., Gatenby, R.A., Maini, P.K., Anderson, A.R.A. Mathematical Model-Driven Deep Learning Enables Personalized Adaptive Therapy. Cancer Res (2024) 84 (11): 1929–1941 (2024). doi: https://doi.org/10.1158/0008-5472.CAN-23-2040.

Ghofraan, A.A., Nirmala, C.K, Kah, T.C., Norfilza, M.M., Reena R.Z., Mohamad N.S., Nor H.A. Predicting Prognosis and Platinum Resistance in Ovarian Cancer: Role of Immunohistochemistry Biomarkers. (2023, January 19). MDPI. Available at: https://www.mdpi.com/1422-0067/24/3/1973. (Accessed 1 September 2024).

Improving Health Equity Through AI. (2024, June 27). Federation of American Scientists. Available at: https://fas.org/publication/improving-health-equity-through-ai/. (Accessed 1 September 2024).

Jiang, Z., Gandomkar, Z., Trieu, P.D. et al. AI for interpreting screening mammograms: implications for missed cancer in double reading practices and challenging-to-locate lesions. Sci Rep 14, 11893 (2024). doi: https://doi.org/10.1038/s41598-024-62324-4.

M Buechel, T J Herzog, S N Westin, R L Coleman, B J Monk, K N Moore. Treatment of patients with recurrent epithelial ovarian cancer for whom platinum is still an option. (2019). NCBI. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8887593/. (Accessed 1 September 2024).

Meta. (n.d.). fastMRI. Available at: https://ai.meta.com/research/impact/fastmri/. 

Mirkhan, A. (n.d.). YOLO Algorithm: Real-Time Object Detection from A to Z. [online] kili-website. Available at: https://kili-technology.com/data-labeling/machine-learning/yolo-algorithm-real-time-object-detection-from-a-to-z. 

Momenimovahed, Z., Tiznobaik, A., Taheri, S. and Salehiniya, H., 2019. Ovarian cancer in the world: epidemiology and risk factors. International Journal of Women’s Health 11, 287-299.

Ovarian cancer by the numbers (no date) OCRA. Available at: https://ocrahope.org/get-the-facts/statistics/#:~:text=Ovarian%20cancer%20accounts%20for%202.5,the%20deadliest%20of%20gynecologic%20cancers. (Accessed 30 August 2024).

Pinto-Coelho, L. (2023). How Artificial Intelligence Is Shaping Medical Imaging Technology: A Survey of Innovations and Applications. Bioengineering, 10(12), p.1435. doi:https://doi.org/10.3390/bioengineering10121435. 

Roke.co.uk. (2020). AI comparative CT scans. [online] Available at: https://www.roke.co.uk/innovative-solutions-for-a-better-future-roke/ai-comparative-ct-scans. (Accessed 24 Aug. 2024]).

Shahzad, M. (n.d.). Cytoreductive Surgery – Ovarian Cancer. Moffitt Cancer Center. Available at: https://www.moffitt.org/cancers/ovarian-cancer/cytoreductive-surgery/. (Accessed 1 September 2024).

Tanabe, K. et al. (2020b) ‘Comprehensive serum glycopeptide spectra analysis combined with Artificial Intelligence (CSGSA-AI) to diagnose early-stage ovarian cancer’, Cancers, 12(9), p. 2373. doi:https://doi.org/10.3390/cancers12092373. 

Tenajas, R., Miraut, D., Illana, C.I., Alonso-Gonzalez, R., Arias-Valcayo, F. and Herraiz, J.L. (2023). Recent Advances in Artificial Intelligence-Assisted Ultrasound Scanning. Applied Sciences, 13(6), p.3693. doi:https://doi.org/10.3390/app13063693. 

Tobia, K., Nielsen, A., Stremitzer, A. (2021). When does Physician Use of AI Increase Liability? Journal of Nuclear Med, 17–21. doi: 10.2967/jnumed.120.256032. 

Wieser, S. (2023). The Future of Sonography: Understanding the Limitations of AI in Ultrasound | 123sonography. [online] 123sonography.com. Available at: https://123sonography.com/blog/future-of-sonography-limitations-ai-ultrasound. 

World Ovarian Cancer Coalition. Ovarian Cancer Key Stats. Available at: https://worldovariancancercoalition.org/about-ovarian-cancer/key-stats/. 

Yanli Wang, Weihong Lin, Xiaoling Zhuang, Xiali Wang, Yifang He, Luhong Li, and Guorong Lyu1, 2024. Advances in artificial intelligence for diagnosing and treating ovarian cancer (Review). 

Zbontar, J., Knoll, F., Sriram, A., Murrell, T., Huang, Z., Muckley, M.J., Defazio, A., Stern, R., Johnson, P., Bruno, M., Parente, M., Geras, Krzysztof J, Katsnelson, J., Chandarana, H., Zhang, Z., Drozdzal, M., Romero, A., Rabbat, M., Vincent, P. and Yakubova, N. (2018). fastMRI: An Open Dataset and Benchmarks for Accelerated MRI. [online] arXiv.org. Available at: https://arxiv.org/abs/1811.08839.