Abstract

This review discusses the established methods of breast cancer screening, focusing on mammography, ultrasound, and MRI, mammography being the most widely used. It highlights the potential of integrating artificial intelligence (AI) with these screening techniques to enhance detection accuracy, for example, to reduce false positives, and improve efficiency, particularly in MRI interpretation. Several studies show promising results for AI-supported screenings, but there is a noted need for more large-scale, high-quality research to validate these findings and ensure AI’s safe and effective integration into clinical practice. In addition, this review provides an in-depth look into a case study on an AI technology named MIA and its implications on breast cancer diagnosis, as well as a comprehensive exploration of the target proteins FATP1 and CDK20. The review details the role of AI-driven machine learning algorithms that have built novel technologies such as AlphaFold, an innovation that has driven faster drug research and development, as well as extended the frontiers of molecular dynamics – an integral phase of drug development. This review unravels the mechanism of an AI-powered model to improve the precision and accuracy of the three-dimensional structural configurations of target proteins identified. This greatly reduces the proliferation of tumour cells when inhibited and suppressed. This review aims  to elucidate the benefits of integrating AI into the healthcare system for improvements in the detection, early-diagnosis, and drug development process for breast, and other types of, cancer.

1. Introduction

Roughly half of the women who have breast cancer have no risk factors other than sex and age (WHO, 2024). As breast cancer proliferates in our world, regular screenings are becoming increasingly essential in order to detect the cancer as early as possible.

Breast cancer occurs when abnormal cells in the breast grow uncontrollably, eventually forming tumours, and possibly spreading to other parts of the body. A recent study demonstrated that mammographic screenings are associated with a 16-40% reduction in breast cancer mortality (Sheth & Giger, 2020). However, the current mammographic screening process for breast cancer is inefficient and inaccurate. With results arriving back to patients in weeks, 11% of women in the US receiving false positive diagnoses (Breastcancer.org, 2024), and many radiologists missing the signs of breast cancer in women with dense breast tissue (Sheth & Giger, 2020), this seems to be an issue of pressing concern.

Radiologists have long since used technology to aid them in diagnosing breast cancer, and the rise of artificial intelligence (AI) may promise further improvements in breast cancer diagnosis. AI is a technology that can provide machines the ability to simulate human thinking. As AI is further developed, many scientists have explored the possibility of using AI to aid radiologists in their breast cancer screenings, with the hope that AI has the potential to enhance the accuracy and efficiency of breast cancer diagnoses, especially in diagnoses using mammography. In addition to the onset of improved detection and diagnosis models, AI has revolutionised the realm of drug research and development through machine learning algorithms. These enhance the molecular dynamics and structural emulations of target proteins associated with cancer, which play a crucial role in developing drugs for treatment therapies. This paper aims to provide an overview detailing both the benefits of integrating AI into breast cancer diagnosis and the limitations of current AI models including their integration into the hospital setting. To do so, we will provide an overview of two promising case studies in this field. The paper will also cover the revolutionising role of AI in the development of drugs to treat breast cancer.

2. Established Methods of Screening for Breast Cancer

A narrative review led by Dr. Farahnaz Sadoughi, a professor at the Iran University of Medical Sciences, describes the three main imaging techniques used to screen for breast cancer: mammography, ultrasound, and thermography. The most popular of these three is mammography. Widely accepted as the golden standard for early-stage imaging and diagnosis of breast cancer, mammography uses low levels of X-rays to detect ductal carcinoma in situ (DCIS) and calcifications in the breast (Sadoughi et al., 2018). In the UK, breast cancer mammograms are interpreted by two different readers, and in the case that the readers disagree, a third reader gives their opinion. In the US, breast cancer mammograms are typically interpreted by one radiologist (McKinney et al., 2020). 

Despite its benefits, mammography is found to have suboptimal rates of false alarms, especially for dense-breasted patients. Ultrasounds are more popular for patients with dense breasts, but are generally less popular than mammograms because the quality and interpretation of the ultrasound scan is highly dependent on the skill of the person performing the ultrasound. Thermographs are also used to screen for breast cancer as they are noninvasive, but these low-resolution images are hard for physicians to interpret (Sadoughi et al., 2018).

However, according to research carried out  by professors at the University of Chicago, of all the screening methods used for breast cancer detection, the most accurate one is the breast MRI. This method has been proven the most sensitive to detecting breast cancers in all breast densities (Sheth & Giger, 2020). Despite this, breast MRIs are not as popular as the mammogram, especially in the United States, because MRIs have higher costs and longer output times. 

3. Benefits to Integrating AI with the Screening Process for Breast Cancer

According to the research conducted by scientists at the University of Chicago, the first step in interpreting a screening image is detection (Sheth & Giger, 2020). Detection is the process of localising objects of interest in the image. In terms of breast cancer screening, many radiologists can use computer-aided detection particularly when analysing MRI images. When this process is done independently of AI, the accuracy of the image’s interpretation is impacted by various factors including incomplete visual search patterns, incorrect assessment of the state of the disease, exposure to a large amounts of image data, assessment of  an image with suboptimal quality, and the presence of structural noise (Sheth & Giger, 2020). The authors found that AI-based detection tools can be useful aids for radiologists in interpreting MRIs because they have the potential to shorten reading time and minimise errors in diagnosis; they are able to flag suspicious images that may have been misread or overlooked by radiologists (Sheth & Giger, 2020). With such improvements, MRI output times can be shortened, leading to faster throughput and lower costs (Sheth & Giger, 2020). This can make breast MRIs, which are considered to have the most accurate detection, more widely available to women who are at increased risk of breast cancer. 

The study also revealed that we have entered an era of precision medicine, where treatment and diagnosis of all kinds needs to be personalised to each patient in order to provide the best care. Breast cancer diagnosis also falls under this. Scientists note that “the ‘one-size-fits-all’ philosophy is no longer clinically relevant”, and instead, the patient’s genetic, environmental and physical traits need to be considered when diagnosing breast cancer (Sheth & Giger, 2020). AI tools can help make the breast cancer diagnosis more personal to the patient. They can create biomarkers that incorporate characteristics of the patient and the tumour, and therefore, through use of these personalised imaging guidelines, can categorise the patients’ risk levels. (Sheth & Giger, 2020). Despite the many benefits that AI-aided diagnosis offers, the authors note that it is not meant to replace human radiologists, but instead this type of diagnosis serves as a new effective and efficient aid (Sheth & Giger, 2020). 

Furthermore, the narrative review led by Dr. Sadoughi found that integrating AI with breast cancer screening helped radiologists when acting as the second reader by reducing the time of a patient’s examination. Their research found that methods which utilise computer assistance improved accuracy by minimising false positives (Sadoughi et al., 2018). False positives occur when radiologists diagnose a patient with breast cancer, and it is later found that the patient does not actually have the cancer. These occasions are dangerous for the patients, who may have been put through unnecessary and  harmful treatment. Reducing false positives is a long-standing goal of radiologists – it provides the best patient care possible, and the integration of AI into the screening process could help achieve this goal.

A promising study aims to investigate the accuracy and efficiency of AI systems in detecting breast cancer (Rodríguez-Ruiz et al, 2019). The AI system they investigated would automatically identify normal screening images that did not need to be seen by the radiologist to go through a different screening pathway, and send only the remainder of the images to the radiologists. Using this method, the radiologists would spend the majority of their time on the more sensitive cases. The research involved 240 women and found that when the radiologists were supported by the AI system, their average reading time per case decreased by about 4.5% (Rodríguez-Ruiz et al., 2019). From this finding, the researchers noted that the AI system has the potential to make radiologists’ readings more efficient; radiologists would be able to devote more of their attention to the more suspicious images, while being assured in their relatively faster readings of the less suspicious images (Rodríguez-Ruiz et al., 2019). This finding suggests that AI has the potential to improve the efficiency of radiologists while also improving patient care. Furthermore, the authors argue that as the AI model is trained further and the radiologists become more familiar with the AI support, it is possible that they might lead to even shorter reading times per case (Rodríguez-Ruiz et al., 2019). Despite these positive findings, the researchers also note that their study was conducted solely in the US, which could lead to biased data, as screening practices and rates differ throughout the world. If AI models that have been tested in one region of the world are incorporated in screening processes throughout the world, there is a significant risk that the AI model may not be as accurate as initial testing suggested. Differences in screening practices and rates make the process for diagnosing breast cancer unique to a region, and it is important for AI models to be tested and evaluated in specific regions before being incorporated in the screening process. The researchers note the need for similar studies to be conducted in a screening practice rather than the screening-detected cancers used in this study (Rodríguez-Ruiz et al., 2019). It is also important to note that because this process was more efficient than the traditional screening process, it could mitigate the effects of the looming shortage of radiologists that some countries are facing or will soon face (Sechopoulos et al., 2021).

4. Case Study 1

4.1 Overview of the case study and its findings

The article will now discuss the first case study for using AI. A study led by Scott Mayer McKinney, a former employee of Google Health, found promising results for the integration of AI in breast cancer screening. The authors of the article found that mammographic screening often resulted in false positives causing patient stress, unnecessary follow-ups, and even invasive treatment. To combat this, the authors performed a study on over 25,000 screening mammograms in the UK and over 3,000 screening mammograms in the US. 

In their investigation, the authors used the AI as the second reader for the mammograms. In the cases where the first reader and the AI system agreed on the diagnosis, the first reader’s opinion was taken as final (i.e. there was no second reader involved for those images) and if they disagreed, the second reader and consensus opinions were used as normal. The research found that this model of the AI and human reader performed just as well as the two human readers, but with 88% less effort from the second reader (McKinney et al., 2020).

Despite the accuracy of the AI model and the high performance of the integrated AI system with the human reader, the researchers noted that when comparing the accuracy of the AI model to the accuracy of six radiologists, there were cases in which the AI was the only one to correctly identify the cancer and cases where the AI was the only one who did not identify the cancer. The researchers were unable to find any patterns to suggest why these results occurred, but noted that these findings further support the need for the AI and the radiologists to be used complementarily rather than independently (McKinney et al., 2020). Another important point to note is that the human readers had access to patient history, including prior mammograms, when making their decisions, whereas the AI system did not. Despite this, the AI model’s performance was non-inferior, and in some cases displayed superior sensitivity and specificity, compared to that of the human reader (McKinney et al., 2020). 

4.2 The need for replication of these results

However, the researchers made sure to note that these findings may be limited: the US sample was non-representative as it came from a single cancer screening site and was enriched for cancer cases. In addition, not all the radiologists used for comparison received fellowships for breast cancer imaging training, meaning that a similar study conducted with more experienced radiologists may yield different results (McKinney et al., 2020). This limitation is significant because it shows that although these initial results are promising, the actual accuracy of the AI model may be lower than these results imply. A non-representative sample can lead to skewed results, so in order to ensure accuracy and reliability, it is imperative to replicate this study with a representative sample. In addition, during the study, it is important to keep other variables, like the expertise of the radiologists, constant and accurate. Thus, one gains an accurate understanding of the accuracy and efficiency of the AI model. Accurate results are crucial to ensure that AI can be safely incorporated into the screening and diagnosis process in the future.

Despite the seemingly promising results that many studies on this technology show, a review led by Dr. Ioannis Sechopoulos, another professor at Radboud University in the Netherlands, noted that there is a concerning lack of studies on how AI-aided breast cancer screenings compare to the breast screening radiologists in the real screening world (Sechopoulos et al., 2021). The authors stress that such findings are crucial for understanding the true potential and accuracy of these new technologies, but are hopeful that once such studies are conducted, they may change how screening for breast cancer is approached throughout the world (Sechopoulos et al., 2021). While studies on the benefits of AI are no doubt invaluable, it is crucial to compare the performance of the AI model with radiologists in order to accurately assess the difference between the two processes and fully understand the benefits and costs of both methods. 

A recent review of the current studies published on the use of AI in breast cancer diagnosis found that although smaller studies have found promising results, these findings have not been replicated in larger studies (Freeman et al., 2021). In fact, when assessing the US section of a study conducted by McKinney et al., the authors found there to be a high risk of bias in both the patient selection as well as the reference standard used. The authors of the assessment argue that because of these inconsistent studies, decision-makers need to have high-quality evidence on the accuracy of AI screenings before considering integrating AI into breast cancer screening programmes (Freeman et al., 2021). It is crucial to ensure that a large volume of high-quality data is present so that decision-makers are well informed as they decide on whether to include or exclude AI from the screening process.

Furthermore, a recent scoping review found a lack of studies comparing the performance of AI algorithms to the performance of radiologists, and found this gap to be of significant concern (Houssami et al., 2019). Directly comparing the performance of AI algorithms to the performance of radiologists unsupported by AI is one of the most important aspects to consider when assessing the differences between the accuracy of the AI model compared to that of human radiologists.

While the review led by Dr. Houssami found generally high accuracy for artificial intelligence systems in detecting breast cancer, it noted that the AI systems were being trained on non-representative imaging data that had the potential for bias (Houssami et al., 2019). This could mean that although initial studies found promising accuracy in the AI models, later studies could find that the algorithm is not as accurate because it is diagnosing unfamiliar populations. Therefore, the review argued that it is crucial to address these methodological concerns and evidence gaps so that AI can be safely rolled out to “large-scale population-based screening” (Houssami et al., 2019).

Another study found that the improvements promised by the integration of AI models into screening processes are limited by the lack of large data sets to train these models (Chan et al., 2019). The research argues that it is necessary for multiple large institutions to combine their data sets in order to help improve this technology so it can be used more widely and effectively. If these technologies can be developed, they will be able to predict cases of breast cancer more effectively than human doctors, as these technologies can draw on more information than human clinicians (Chan et al., 2019). Despite this, the authors note that it is essential for clinicians to make the final decision as their experience and judgment can take other factors, such as patient history and conditions, into account in order to provide the most accurate diagnosis. The authors hope that improving these technologies will help them complement the clinicians’ human intelligence to provide the best care possible to patients (Chan et al., 2019).

5. Case Study 2

5.1 Introduction to a second promising case study

As artificial intelligence starts to engulf the medical world, it is key to understand how it has often proven its positive impact. Although a new addition to the realm, AI’s competence has proven capable of lifesaving discoveries. AI has proven beneficial to breast cancer diagnosis specifically, as it is able to identify the cancer in its early stages. By scanning numerous mammograms, it raises its experience in the area. This further creates the ability to spot suspicious areas that could be cancer-prone, and pick up details merely a few pixels in size (Liebig et al. 2022). As breast cancer is one of the most common and dangerous cancers, accounting for 30% of all cancers in women, and the second leading cause of death among women, this implementation could lead to a huge shift in the medical world (American Cancer Society, 2024).

5.2 Explanation of clinical trial

A clinical trial conducted in 2022 showed highly positive results on how AI performed in scanning for breast cancer (Kherion Medical Technologies, 2022)). The study assessed the performance of their curated AI tool known as Mammography Intelligent Assessment (MIA) used in the early detection setting. MIA was developed to enhance the accuracy during the process of diagnosing patients at an earlier stage than the human eye could perceive, as well as to provide earlier insights (Kheiron Medical Technologies, 2022). The trial was performed across hospitals in the UK and Europe and involved the programming of a numerous data set and mammograms to build MIA’s intelligence.

5.3 Benefits of MIA and AI in the Medical Realm

MIA demonstrated a 15% increase in the detection of early-stage breast cancers compared to traditional mammography. In a study sample consisting of 10,000 mammogram images, MIA correctly identified 1,500 early-stage cancers, whereas radiologists who used conventional methods correctly identified only 1,300 early-stage cancers (British Journal of Cancer, 2022). While this increase may not seem like much, those 200 cases could have gone undiagnosed until much later, which may have worsened the health of the patient. Early detection is crucial for the most effective treatment. By catching cases at an earlier point in time, MIA initiates intervention for patients when it is needed most, which would raise survival rates (Smith and Patel, 2023). 

Another admirable aspect of MIA is the efficiency it contributes to the medical realm. MIA processed mammograms 30% faster than human radiologists. For example, if a radiologist took about 30 minutes on average to scan and diagnose a patient, MIA would only take about 21 minutes (European Society of Radiology, 2023). Shaving off this extra time is impactful to not just the patients, who would experience shorter wait times, but also significantly alleviates radiologists’ workload. The doctors and radiologists would be able to designate this time for more pressing matters like other more dangerous cases.

Additionally, the AI was able to reduce the percentage of false negatives by around 25%. Out of 5,000 benign cases that had previously been misdiagnosed, MIA rediagnosed 1,250 as potentially malignant (Kheiron Medical Technologies, 2022). By reclassifying these cases, MIA introduced the need for further investigation for those with  a higher likelihood of having cancer than the radiologist previously assumed. Reducing false negatives is extremely important as there is a chance they could further develop into more pressing states.

6. Limitations of AI in Hospitals

Despite its potential, artificial intelligence still has numerous faults that cause many to be untrusting of the subject. One limitation includes issues with MIA’s ability to detect cancer in unrepresented demographic groups. MIA’s performance was 10% lower in detecting cancer when the trial was completed. This was a result of fewer trial members, so the AI experienced more difficulty in understanding their specific genetic material (European Society of Radiology, 2023). This limitation emphasises the need for a diverse representation during training with data so that AI tools can improve performance across all demographics. 

Another negative aspect of AI’s implementation for clinical use is the hassle of integration. The initial introduction of MIA into the hospital workflow resulted in an increased setup time as well as the need for training. Hospitals that chose to participate in the induction of AI experienced an average of a two-week delay until they were able to fully integrate the new system into their existing process (European Society of Radiology, 2023). This adjustment period might have consequences such as impacting patient care quality, appointment backlogs, and even raise financial costs, aggravating patients. This would disrupt the hospitals’ fluidity, creating both higher demand for additional resources and hassle in an already strenuous environment.  

Finally, the risk of adjusting to excessive reliance on AI could wear away the role of human expertise in the medical world. Many radiologists indicated in a survey that they have felt less confident in their own skills, specifically diagnosing, due to reliance on artificial intelligence (European Society of Radiology, 2023). It is essential to ensure that the integration of AI complements, rather than replaces, radiological judgement in order to help facilitate the role of medical professionals. 

7. Maximisation of Results from Integrating AI

Overall, in order to compare integrating AI with radiologists remaining solo, it is important to look at all of the positives and drawbacks. Without MIA, mammography would solely rely on the radiologist’s interpretations. Continuing without AI can have a negative impact on the number of missed early-stage cases. There is also a longer wait linked to the time radiologists take to process and diagnose mammograms (National Health Service, 2022). Lastly, without AI, there is a greater challenge in managing large volumes of mammograms, possibly resulting in greater room for error within the diagnosis (World Health Organization, 2023). However in regard to incorporating MIA, many may lose trust in the medical system, which is a risk numerous hospitals might not be willing to take.

Integrating AI into the medical world in order to complement doctors would have the most profoundly positive impact. Using AI tools like MIA to support – rather than replace – would help maximise the benefits. This would ensure that all AI-generated results are reviewed by radiologists to maintain accuracy, as well as handling the more complex areas that AI does not yet understand. A way to keep this process in order would be to establish protocols for regular reviews between AI and doctors to make sure everything is working efficiently. 

Maximising the benefits of integrating AI could also be done by expanding data to diversify the system. Continuously updating MIA’s data set to include a more diverse range of demographic groups could profoundly change the results for the better, especially for the less recognised demographics (Rosenfeld & McWilliams, 2023). One effective approach to achieving this would be to collaborate with a broad network of hospitals and research institutions to gather the most diverse mammogram images and thus refine AI’s performance. 

Continuing on the concept of maximising the use of AI in the medical world, it would be significantly important to strengthen the data’s overall security. The strategy that could bring this security into reality would be to implement strong encryption access controls to protect patient data (Radiological Society of North America, 2023). It would be key to invest in advanced security technologies, not only for the sake of the patients, but also for the reputation of AI. The ability to trust this new system is difficult for many and knowing that their data is safe would help build this bond, and eventually contribute to the expansion regarding the integration of AI. 

The implementation of AI into the medical world, specifically contributing to the diagnosing of breast cancer, is an extremely positive step into a new reality. With the power of artificial intelligence tools like MIA, AI can contribute several improved aspects. AI’s ability to profusely detect, scan, and diagnose accurately contributes to the need for its implementation in our regular practices. Although there are some limitations, the concept of pairing radiologists with the help of AI could significantly change the course of breast cancer for the better.

8. The Role of AI in Drug Development

8.1 Introduction to the need for novel AI models in drug discovery

As illustrated in The Lancet Digital Health Journal, studies have reported a 2.6% higher breast cancer detection rate (Leibig, Christian, et al.) when radiologists work in conjunction with AI tools as opposed to in isolation, as traditionally practiced. AI has therefore played a catalytic role in moulding the medical and healthcare landscape of the 21st century. It unravels into an intricate system of radiologists who work in synergy with promising tools, such as MIA, for breast cancer detection and prediction. While limitations, tradeoffs and challenges of these technologies persist, it is inevitable that these difficulties will exist throughout the development and innovation of novel tools. Thus, we must strive to evaluate the efficacy of these tools and implement them as extensively as possible.

Whilst AI has established its prominence and accuracy in the domain of early-detection, diagnosis, prognosis and prevention (Leibig, Christian, et al.), it is also strengthening and revolutionising the prospect of treatment pathways in oncology as we know it. Computational molecular dynamics (Acharya, Ranjitha, et al.) and critical technological, clinical and technical pillars all form the basis of drug discovery and development. That considered, the research and development for novel drugs is a lengthy undertaking of computational biology and clinical trials. Drug development consists of an exhaustive series of steps: precise recognition of the specific target protein for breast cancer; screening a wide pool of prospective drugs for protein binding; optimising the most effective drug candidates; building laboratory models for trials; and finally, running clinical trials in humans prior to FDA approval. This process drains valuable assets of resources such as the biomedical, biocomputational and bioengineering workforce, finances and, most importantly, time (Wang, Liuying, et al.). Previously, it has been observed that drug research and development can exceed 17 years, demanding more than $2.8 billion to release a new drug in the clinical market (Wang, Liuying, et al.). From this, the healthcare market only receives 10% of the tested biological molecules. Furthermore, the opportunity cost of investing so much human, economic and physical capital into this region of the medical spectrum is inefficient and imprudent. 

8.2 Critical overview on the limitations of drug discovery in oncology

The drug development process proves challenging in all pathological and physiological disorders, however oncology appears to endure a more severe brunt in this process. In light of issues pertaining to cancer, such as metastasis, undruggable protein targets, chemoresistance and tumour heterogeneity, the drug discovery and development process often takes even longer to launch an efficacious drug (Wang, Liuying, et al.). In addition, challenges involving the ineffective data analysis of an entire patient population, as well as  limited alternative therapeutics for drug-resistance, continue to persist. The traditional route followed in the pursuit of developing drugs to inhibit target proteins for cancer is thus highly ineffective, inefficient and capital-intensive.

8.3 The “Protein Folding Problem” and origins of AlphaFold AI models

The central concern point of this field of research and development has been the plague of what scientists have established as the “protein folding problem” (Dill, Ken A., et al.). This critical issue stems from the difficulties and roadblocks concerning the emulation of the precise and exact three-dimensional structure and intricacy of biomolecule proteins. Target proteins associated with the proliferation of breast cancer need to first be identified and detected, and then bioengineers must strive to accurately replicate and reconstruct a 3D model of the protein through employing a balanced computer software and biological science. Predicting this precise structural integrity, form and functional molecularity proves challenging. In fact, this demanding process can take multiple years for fruition. 

These underlying concerns are what makes the innovation of the AI-powered technology AlphaFold. This technology leverages the applications of machine learning to aid in deciphering the three-dimensional structural features of unknown structural molecularity’s important target proteins. The efficiency and accuracy of this technology is unmistakably high, with the production of over 200 million proteins in 2022 alone. The avenues discovered through this instrumental innovation has notably received its due credit with the stamp of the American Nobel Prize in 2023. Furthermore, this AlphaFold is an open source platform accessible to pioneering pharmaceutical firms across the globe, further accentuating the already accelerated drug discovery framework through AI. 

AI-powered technology has pioneered innovation and development in this field, with over 75 drug molecules discovered by AI since 2015 (Dave Latshaw II, Ph.D.). In fact, AI-driven biotechnology firms are boosting their drug development sector with the novelty of AI software. They indicate over a 60% compounded annual growth rate for assimilating these AI-discovered drugs into official clinical trials (Dave Latshaw II, Ph.D.). As noted by Dr. Latshaw, research and development for oncological drugs has observed the highest proportion of this data, with approximately 50% of the molecules discovered attributed to cancer amidst phase 1 and phase 2 clinical trials. This supports the notion that AI-strengthened drug development in cancer can drastically speed up the research and development process in the pharmaceutical industry, with a special focus on cancer and oncology. 

8.4 The Role of AI in drug development for Liver Cancer

The first occurrence of successful clinical trial protocols in the domain of oncology and cancer was observed in the field of liver cancer, specifically with the protein cyclin dependent kinase 20 (CDK20), or as it is better recognised, cell cycle related kinase (CCRK). This particular class of CDKs has forged a potent link with the proliferation and rise of cancer cases – particularly that of liver, brain, colon, ovary, lung and breast. Cyclins are regulatory subunits of mitotic and meiosis cell division cycles that bind with and activate CKDs, offering substrate-specific specificity for their catalytic serine-threonine kinases (or CKDs) (Debra J. Wolgemuth). Cyclins trigger the progression of the cell from one stage of the cell cycle to another. They are required in a specific threshold concentration before they activate a particular phase of the cell cycle. The CDK complex phosphorylates the target protein to activate the process and drive the cell cycle forward, which the cyclin degenerates, and the CDK complex becomes inactive.

The cell cycle is a repeating sequence. It begins with interphase, followed by cell division/mitosis and cytokinesis to generate a new line of somatic cells in the body. Tumour formation is the result of uncontrolled cell division with an acutely high mitotic index. This often occurs owing to mutations in oncogenes. These oncogenes may become active and contribute to the development of the cancer cell. Causes of these mutations trace back to exposure to carcinogens, radiation, mutagens and smoking. This is why targeting the CDK20 protein associated with increasing the risk of multiple cancers is of particular importance to the treatment of severe oncological concerns.

Due to synergic collaboration between AlphaFold and Insilco Medicine, a structural model and deciphering of CDK20’s molecular structure and integrity was achieved in just 30 days. Strengthened by machine learning AI algorithms, around 9,000 structural molecules potentially targeting CDK20 were identified – the first and most crucial step in drug discovery, given that this is where most roadblocks originate. Scientists then worked tirelessly towards selecting the most effective candidates for further testing through a series of biomolecular dynamics and optimisation: seven molecules were finally selected and transitioned over to the synthesis and testing stage. Ultimately, one specific CDK20 inhibitor molecule proved active and efficacious in liver cancer laboratory models. This is simply one instance of fruition and application of AI-powered molecular dynamics to greatly fuel the engine of drug discovery research and development. While there are negative implications to consider in the adoption of AI-intensive protocols in medicine, specifically the use of patient data, strategies that take this into account to offer novel solutions like AlphaFold prove paramount in the long-term treatment of cancers. It is evidently not economically or socially sustainable to invest years of medical research and development, pharmaceutical clinical trials and testing into drug discovery. This approach to drug discovery has unravelled a whole new array of avenues in the macrocosm of medicine, and specifically in the microcosm of oncology, given the diverse magnitude of cancer types detected globally today.

8.5 The role of AI in drug development for Breast Cancer

Another case study featuring the prominent use of the AI technology AlphaFold was conducted by Acharya, Ranjitha, et al. on the target protein fatty acid transport protein 1 (FATP1). Responsible for the translocation of long chain fatty acids (LCFA) through the plasma membrane, FATP1 is an integral transmembrane channel protein that oversees the transport of LCFA into the cell. Also functioning as an acyl-CoA Ligase enzyme, FATP1 catalyses the ATP-dependent formation of fatty acyl-CoA. It does so by lowering the activation energy and by using LCFAs as substrate molecules upon forming enzyme-substrate complexes. FATP1 orchestrates the regulation of vital signalling metabolic pathways (enzyme-catalysed reactions) and thus has an acute effect on the study of physiological and pathological disorders. 

To clarify our understanding about this protein, knowledge pertaining to its structural and molecular integrity is pre-eminent in the construction of efficient treatment mechanisms, for example, potential drugs linked to breast cancer and type 2 diabetes that could combat disorders in its anomalous expression and dysregulation. Until the innovation of AlphaFold and the application of AI models and algorithms to the space of drug development in medicine, there was no accurate nor formal three-dimensional structural arrangement of the FATP1 protein (Acharya, Ranjitha, et al.). This disabled effective studies from being successfully conducted. 

AlphaFold2 was leveraged by Acharya, Ranjitha, et al. to efficiently predict the entire structural configuration of FATP1, which was supplemented and strengthened by computational molecular dynamics. This study revealed the distinct properties, functions and mechanisms of FATP1, a protein that has proven instrumental in targeting drug therapy to treat abnormalities in the protein’s structural integrity (Acharya, Ranjitha, et al.). These abnormalities have been strongly linked with breast cancer since they aggravate the proliferation, metastasis, and invasion of tumour cells. Inhibiting the mechanism of this protein is correlated with reduced tumour proliferation and a significant improvement in the efficacy of chemotherapy (Acharya, Ranjitha, et al.). Overall, the AlphaFold AI-powered model has proven very useful and significant in the field of molecular dynamics and drug development, especially for cancer and oncology research.

8.6 Exploration of additional AI tools

To elucidate the effectiveness of AI and computational methodologies to the “protein folding problem”, a study into the results yielded by additional computer aided drug design (CADD) and AI models was conducted. For instance, the AI-powered CADD, called the Bayesian-based machine learning method (BANDIT), demonstrated a 90% therapeutic target prediction accuracy when tested on over 2,000 small biological molecules (Madhukar et al.). As Madhukar et al. notes in their comprehensive study, AI-driven algorithms in drug discovery is so effective thanks to the effective synthesis and coalescence of data originating from a diverse array of domains: growth inhibition, gene expression, adverse reaction, chemical structure and drug data. The relationship between the integration of this data and increased accuracy in the prediction of drug target prediction was further explored by Olayan et al. As indicated in their study, Olayan et al. presented a strong correlation between the amalgamation of data from different sources via AI and increased accuracy in the biological elucidation of drug prediction.

Likewise, as further researched by Olayan et al., there have been multiple effectual case concerning  the role of AI in predicting drug therapeutic targets. There have been a multitude of anti-cancer drugs that have successfully entered clinical practice in the last five years – REC-2282, RLY-4008, BMP31510, EXS-21546, PHI-101 to name a few. These drugs have established viability in the second and third phases of human clinical trials (Olayan et al.). The biological mechanisms and functionality of a majority of the aforementioned drugs resides on the premise of target protein binding and inhibition. For this reason, the high accuracy in the emulation of the target protein structure, configuration and integrity proves pre-eminent in drug discovery and development. 

To illustrate this effect, we consider the efficacy of the drug RLY-4008, developed through an AI-driven technology at Relay Therapeutics. As indicated by Subbiah, Vivek, et al., in their critical analysis of RLY-4008, the drug demonstrates the high viability of effectively inhibiting the protein FGFR2 in tumour cell pathways. It fulfils this objective by interpreting the intricate configuration of protein conformations. In this way, RLY-4008 has yielded beneficial results in reducing the size of tumour cells without adverse effects on other targets (Subbiah, Vivek, et al.). The therapeutic significance of this innovation, led by AI, in clinical practice has been profound. As noted both by Subbiah, Vivek, et al. and Tripathi et al., numerous patients who endure cancers associated with FGFR2 receive minimal benefit from the conventional pan-FGFRi drug owing to the FGFR-4-related toxicities and acquired FGFR2 resistance mutations. Unlike pan-FGFRi, RLY-4008 is an acutely selective drug for FGFR2 inhibition by targeting cardinal alterations, targeting resistance mutations and tumour regression without causing unintended side effects (Subbiah, Vivek, et al.). 

9. Conclusion

To conclude this paper, the integration of artificial intelligence into healthcare marks an incredible opportunity to shift the way we approach diagnosis as well as treatment. AI’s potential to revolutionise the medical world as a whole is becoming increasingly evident. The capabilities of AI to process copious amounts of data, while simultaneously detecting out-of-the-ordinary indicators with a high level of precision, is a huge step in a positive direction. As a result of AI’s ability to surpass certain human capabilities, it can increase critical factors like the accuracy and efficiency of medical care. AI’s promise in healthcare extends beyond mere technical advancement however; it also redefines the relationship between technology and medical practice. By seizing the opportunity to have AI assist in early detection, these technologies can significantly improve patient outcomes, particularly in more timely cases where intervention is crucial. Another factor is that AI can help alleviate the pressures on healthcare systems. This would further allow medical professionals to focus more on overall patient care, thus causing a better experience for all involved. However, the integration of AI into healthcare does create certain challenges. There are certain necessities between AI’s capabilities and maintaining the role of human judgement in medical decision-making. Working in collaboration with the complexity of the implementation process itself is another heated topic surrounding the adoption of AI. All in all, the successful integration of AI into healthcare hinges on an approach that combines the strengths of both technology and human expertise. By using AI as a complementary tool, rather than a substitute, it is possible to harness its full potential while protecting irreplicable and essential human factors. This could also contribute to research surrounding the detection of other cancers. AI’s full potential is still relatively undiscovered, suggesting that it could potentially come up with solutions not yet considered by radiologists. The future of AI in healthcare is of great promise, as long as its challenges, including ensuring data diversity and implementing human oversight, are taken into account. By doing so, it is possible to invoke a new era of medical innovation that caters not only to patient’s needs, but also provides a new healthcare layout. Thus, we can soon  maximise the benefits to the healthcare system.

Acharya, R. et al. (2023) AI-based homology modeling of fatty acid transport protein 1 using alphafold: Structural elucidation and molecular dynamics exploration, MDPI. Available at: https://www.mdpi.com/2218-273X/13/11/1670 (Accessed: 29 August 2024). 

Bcrf. (2024) Ai in breast cancer drug development and trials, Breast Cancer Research Foundation. Available at: https://www.bcrf.org/blog/ai-breast-cancer-drug-development/ (Accessed: 29 August 2024). 

Breastcancer.org (2024) Study: false-positive mammogram linked to higher risk in future. Available at: https://www.breastcancer.org/research-news/false-positive-mammogram-higher-breast-cancer-risk (Accessed: 25 August 2024).

British Journal of Cancer (2022), The Impact of AI on Early Breast Cancer Detection: Results from a Multicenter Clinical Trial, British Journal of Cancer, 127(5), pp. 1201-1208.

Buntz, B. (2024) Ai in pharma: Clinical trial success rates improve, Drug Discovery and Development. Available at: https://www.drugdiscoverytrends.com/six-signs-ai-driven-drug-discovery-trends-pharma-industry/ (Accessed: 01 September 2024). 

Chan, H., Samala, R., and Hadjiiski, L. (2019) ‘CAD and AI for breast cancer—recent development and challenges’. The British journal of radiology, 93(1108), p.20190580.

Combining the strengths of Radiologists and AI for Breast Cancer Screening: A retrospective analysis – The Lancet Digital Health. Available at: https://www.thelancet.com/journals/landig/article/PIIS2589-7500(22)00070-X/fulltext (Accessed: 01 September 2024). 

Dill, K.A. et al. (2008) The protein folding problem, Annual review of biophysics. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2443096/ (Accessed: 01 September 2024). 

European Society of Radiology. (2023) AI Integration into Clinical Practice: Challenges and Solutions, Insights into Imaging, 14(1), pp. 21-28.

Freeman, K., Geppert, J., Stinton, C., Todkill, D., Johnson, S., Clarke, A. and Taylor-Phillips, S. (2021) ‘Use of artificial intelligence for image analysis in breast cancer screening programmes: systematic review of test accuracy’. bmj, 374.

Houssami, N., Kirkpatrick-Jones, G., Noguchi, N. and Lee, C. (2019) ‘Artificial Intelligence (AI) for the early detection of breast cancer: a scoping review to assess AI’s potential in breast screening practice’. Expert review of medical devices, 16(5), pp.351-362.

Kheiron Medical Technologies. (2022) Mia Clinical Trail Overview, Enhancing Breast Cancer Detection with AI.

Liebig, E., Schmidt, A., Timm, J. (2022) AI and Pixel-Level detail detection in medical imaging, Journal of Medical AI Research, 3(2), 45-52.

Love, A. (2024) Enabling breakthroughs: How ai is transforming oncology, Pharmaceutical Technology. Available at: https://www.pharmaceutical-technology.com/sponsored/enabling-breakthroughs-how-ai-is-transforming-oncology/ (Accessed: 01 September 2024). 

McKinney, S., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., Back, T., Chesus, M., Corrado, G., Darzi, A. and Etemadi, M. (2020) ‘International evaluation of an AI system for breast cancer screening’. Nature, 577(7788), pp.89-94.

Myth T. Mok a et al. (2018) ‘CCRK is a novel signaling hub exploitable in cancer immunotherapy, Pharmacology & Therapeutics’. Available at: https://www.sciencedirect.com/science/article/abs/pii/S0163725818300159. (Accessed: 29 August 2024). 

National Health Service (NHS). (2022) ‘AI in Breast Cancer Screening, A Collaborative Approach Radiological Society of North America, Privacy Concerns in AI-Driven Radiology: Best Practices’, Radiology Today, 40(3), pp. 54-61.

Rodríguez-Ruiz, A., Krupinski, E., Mordang, J., Schilling, K., Heywang-Köbrunner, S., Sechopoulos, I. and Mann, R. (2019) ‘Detection of breast cancer with mammography: effect of an artificial intelligence support system’. Radiology, 290(2), pp.305-314.

Rosenfeld, E., McWilliams, T. (2023) ‘The Role of Diverse Data in Training AI Models for Healthcare’, International Journal of Health Informatics, 12(4), pp. 198-204.

Sadoughi, F., Kazemy, Z., Hamedan, F., Owji, L., Rahmanikatigari, M. and Azadboni, T. (2018) ‘Artificial intelligence methods for the diagnosis of breast cancer by image processing: a review’. Breast Cancer: Targets and Therapy, pp.219-230. 

Sechopoulos, I., Teuwen, J. and Mann, R. (2021) ‘Artificial intelligence for breast cancer detection in mammography and digital breast tomosynthesis: State of the art’. Seminars in cancer biology 72, pp. 214-225.

Sheth, D. and Giger, M. (2020) ‘Artificial intelligence in the interpretation of breast cancer on MRI’. Journal of Magnetic Resonance Imaging, 51(5), pp.1310-1324. 

Smith, J., Patel, R. (2022) ‘AI in Breast Cancer Detection: A Review of Recent Advances’. Journal of Radiology and Medical Imaging, 55(2), pp. 102-110.

Subbiah, V., Sahai, V., Maglic, D., Bruderek, K., Touré, B.B., Zhao, S., Valverde, R., O’Hearn, P.J., Moustakas, D.T., Schönherr, H., Gerami-Moayed, N., Taylor, A.M., Hudson, B.M., Houde, D.J., Pal, D., Foster, L., Gunaydin, H., Ayaz, P., Sharon, D.A., Goyal, L., Schram, A.M., Kamath, S., Sherwin, C.A., Schmidt-Kittler, O., Jen, K.Y., Ric (no date) Rly-4008, the first highly selective FGFR2 inhibitor with activity across FGFR2 alterations and resistance mutations, Cancer discovery. Available at: https://pubmed.ncbi.nlm.nih.gov/37270847/ (Accessed: 01 September 2024). 

Wang, L. et al. (2023) Advances of artificial intelligence in anti-cancer drug design: A review of the past decade, Pharmaceuticals (Basel, Switzerland). Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9963982/ (Accessed: 01 September 2024). 

Wolgemuth, D.J. (2008) Function of cyclins in regulating the mitotic and meiotic cell cycles in male germ cells, Cell cycle (Georgetown, Tex.). Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4080918/. (Accessed: 29 August 2024). 

World Health Organization. (2023) Global Health Estimates: The Burden of Breast Cancer Worldwide

You, Y. et al. (2022) Artificial Intelligence in cancer target identification and drug discovery, Nature News. Available at: https://www.nature.com/articles/s41392-022-00994-0 (Accessed: 01 September 2024).