February 7, 2024
One important development in the fight against cancer is the application of artificial intelligence (AI) in oncology. The ability of AI to analyse data quickly presents a novel method for identifying and treating this complicated illness. Artificial intelligence has great promise for improving patient outcomes through personalised treatment regimens and enhanced diagnosis accuracy.
This article examines how artificial intelligence is changing cancer care, outlining its uses, advantages, and difficulties. AI is transforming the oncology field and opening doors for more tailored, focused, and efficient cancer treatment plans.
The Evolution of AI in Oncology
Artificial Intelligence (AI) is fundamentally transforming our existence, making it crucial to grasp its progression and accomplishments to inform future development strategies. This is particularly relevant in oncology and associated areas, where AI is unveiling significant new possibilities for enhancing the care and management of individuals with cancer.
In 1950, Alan Turing pioneered the concept of utilising computers to simulate intelligent behaviour and critical reasoning. Today, AI is an advancing and swiftly developing technology with applications across various scientific disciplines, including those focused on cancer patient care.
Numerous AI technologies have been sanctioned by the US Food and Drug Administration (FDA) for application in oncology, especially within the field of radiology. In 2021, AI-enabled devices authorised for oncology were predominantly utilised in radiology (54.9%) and pathology (19.7%), with breast cancer (31.0%) being the most frequent cancer type for which these devices were employed, compared to other types of cancer.
Current Applications of AI in Oncology
Breast radiologists are increasingly utilising AI for a variety of tasks, including assessing breast density, evaluating mammogram quality, categorising mammograms based on the risk of breast cancer (low, intermediate, or elevated), detecting individuals at risk for atherosclerotic disease, and identifying breast cancers. Furthermore, AI could be applied in breast ultrasound exams to offer an additional layer of analysis by determining if the findings from a breast ultrasound are likely indicative of cancer.
AI has shown effectiveness in screening and diagnosing colorectal cancer. In 2021, the FDA approved GI Genius for marketing, making it the first device that utilises AI to assist in detecting lesions during colonoscopies.
AI is being employed in some pathology laboratories to analyse digital pathology slides and enhance cancer diagnosis, following extensive research in the field. AI is also playing a role in developing new cancer drugs, with some companies using AI to identify novel and safe oncology targets, while others employ AI, sometimes in conjunction with molecular modelling, to create new drug candidate molecules.
Additionally, AI is improving the efficiency of radiation treatment planning and aiding in tumour and organ contouring, thereby accelerating the initiation of therapy and enhancing the effectiveness and safety of radiation treatment.
For supportive care, AI is utilised to evaluate remotely monitored patients’ self-reported symptoms and vital signs. Natural language processing and machine learning models can generate alerts for adverse health trends.
The Potential of AI in Oncology
AI’s application in oncology is expanding, with research suggesting promising future uses. Among these, breast imaging emerges as a key area ripe for AI exploration, potentially offering methods to assess near- and long-term breast cancer risks more accurately than relying solely on family history or as an adjunct tool alongside it.
Furthermore, the use of AI in detecting high-risk pancreatic cancer patients through abdominal imaging and analysis of longitudinal electronic health records is under investigation. This approach could lead to the development of screening programs aimed at high-risk individuals, potentially enabling early detection in diseases known for their high mortality rates.
Additionally, AI is set to revolutionise the way biomarkers and the molecular characteristics of cancers are assessed. This advancement could greatly inform prognosis and treatment choices without the need for invasive biopsies, reducing the waiting time for results in some instances. The potential for AI to facilitate this through less invasive means, such as blood tests or ‘virtual biopsies’ using MRI radiomics, is particularly exciting. For example, a deep learning model might analyse a brain MRI to identify specific genetic mutations in a tumour, offering a glimpse into a future where diagnosis and treatment planning are significantly more patient-friendly and efficient.
Large language models (LLMs) are emerging as a promising field with potential implications for oncology. There’s preliminary evidence suggesting that certain models developed by OpenAI might accurately respond to medical inquiries, though a more rigorous validation with high-quality datasets is required. Medical centres, with their vast reserves of oncology data, could potentially harness this data to train sophisticated LLMs. These models could then analyse lab results, scans, and medical histories to offer highly personalised treatment suggestions.
Recent research supports the utility of these models in oncology. A study found that ChatGPT could deliver accurate answers to 88% of inquiries about breast cancer screening and prevention, as verified by breast radiologists. Additionally, research has shown that chatbots can reliably answer questions across various cancer types, although there are still hurdles to overcome. For instance, while chatbots generally provided reliable information, it wasn’t always actionable and often required a higher level of literacy to comprehend.
Further examination revealed that chatbots’ advice on cancer treatments occasionally diverged from the National Comprehensive Cancer Network (NCCN) guidelines. Despite the chatbot offering at least one treatment suggestion for nearly all inquiries about breast, prostate, and lung cancer, some recommendations did not align with NCCN guidelines, and a portion of the responses were inaccurately fabricated.
Another innovative AI application being explored is the use of natural language processing to predict cancer patient survival outcomes based on initial oncology consultations. This approach has shown promise, indicating that AI models can forecast survival rates using just the data from a patient’s first visit to an oncologist.
Challenges and Obstacles
To fully harness AI’s potential in clinical oncology, several challenges and limitations need addressing. It’s crucial to develop AI models and AI-enabled medical devices using diverse data sets to accurately reflect the patient populations they will serve in clinical settings. Failure to do so could result in biassed outcomes and widen healthcare disparities.
A significant obstacle in research is the reluctance of medical centres to share data. Strategies to normalise and encourage data sharing are essential, possibly requiring innovative training algorithms that allow for local model training with subsequent merging and accuracy verification across different centres.
Despite the advancements and potential of AI in medicine, many models have yet to be tested in clinical environments. An AI model’s true value in oncology comes from its application in clinical care, where it should offer benefits to clinicians and enhance patient and health system outcomes. The practicality, acceptance, and safety of these models are yet to be fully established, necessitating continuous evaluation to gauge their effect on healthcare outcomes.
Current implementation hurdles that limit AI models’ integration into clinical settings include difficulties in incorporating AI into electronic health record systems, the lack of user-friendly interfaces, and the high costs associated with training some of the more recent models. Addressing these issues is critical to avoid limiting competition and access, suggesting a need for subsidised computing power or more resource-efficient training strategies.
Furthermore, establishing “guidelines and guardrails” is vital for ensuring AI’s safety and preventing harm. There must be transparency regarding the data sets used in AI models to understand their limitations and identify appropriate patient cohorts clearly. Additionally, maintaining patient confidentiality is paramount.
The advent of generative AI and the development of large language models represent a transformative moment in oncology, potentially having a profound impact on cancer research and the delivery of care.
As we conclude our exploration of Artificial Intelligence (AI) in oncology, it’s clear that AI is not just a futuristic concept but a present reality transforming cancer care. Through its capacity to analyse vast amounts of data with unprecedented speed and accuracy, AI is ushering in a new era of personalised treatment and enhanced diagnostic precision. From improving radiology and pathology to developing new cancer drugs and refining treatment plans, AI’s applications in oncology are vast and varied.
Yet, the journey ahead is not without its challenges. To realise AI’s full potential, the medical community must navigate ethical considerations, data sharing hesitations, and the integration of AI into clinical practice. Ensuring diverse data sets to avoid bias, overcoming technological and financial barriers to AI implementation, and continuously evaluating AI models for safety and effectiveness are critical steps toward optimising AI for clinical use.
As we stand on the brink of what could be a revolution in cancer care, it’s essential to continue pushing the boundaries of what AI can achieve in oncology. By harnessing AI’s power responsibly and innovatively, the fight against cancer can be significantly advanced, leading to better outcomes for patients worldwide. The promise of AI in oncology is immense, and with continued research, collaboration, and ethical consideration, its full potential can be unlocked, marking a new chapter in the relentless battle against cancer.
Published on 07-02-2024