top of page

West Oncology 2024: Where is AI Going in Oncology?

Updated: Mar 7

Presentation by Adam P. Dicker, MD, PhD, FASTRO, FASCO, Thomas Jefferson University

At the 2024 West Oncology Conference in Memphis, Tennessee, Dr. Adam Dicker from Thomas Jefferson University presented on using artificial intelligence (AI) methodologies to solve complex problems in oncology.


Dr. Dicker began by reviewing some evidence for inherent biases that may exist when physicians treat patients. For example, just as buyers may be swayed to purchasing something that is $ 7.99 as compared to $8.00, physicians may be less inclined to offer lifesaving treatments such as coronary artery bypass grafting (CABG) procedures to individuals who are 80 years and two weeks old, as compared to those 79 years and 50 weeks old.  Such “left-digit bias” may also apply to those patients with rectal cancer, as one study found a 36% decrease in the likelihood of physician adherence to guideline-based treatment for rectal cancer with each decade of increase in age.


Dr. Dicker described the overall progression of the AI hierarchy over time, evolving from the 1950s era “if-then” models and following a decision tree to the early 2010s (“deep learning”), with prediction and classification of information based on new data, through 2018 and beyond (“foundation models”), which can generate new content and perform different tasks without the need for new data or retraining.


Radiation Oncology


Dr. Dicker described the use of AI to address several unmet needs in radiation oncology, for example, reducing the amount of collateral damage to normal tissues in patients undergoing radiotherapy, integrating multiple sources of data (“multi-omic” data) for a more personalized treatment approach, and to better support clinical decision making when there are multiple treatment options.  He noted there are also opportunities to improve the quality and safety of radio-oncology by reducing the substantive need for human resources and the necessity to check everything, as well as by assimilating and aggregating the large amount of data generated for each patient.  Lastly, there is an opportunity to reduce toxicities, including financial ones, by repurposing a large amount of electronic medical records (EMR) and imaging data that often go unused.  He cited some prospective studies underway at his institution using AI and 4D computed tomography (CT) scanning techniques to reduce damage to normal tissues in lung cancer patients receiving chemotherapy and radiation.


Reducing Medical Errors and Cognitive Burden


In colorectal cancer screening, Dr. Dicker noted evidence from a recent meta-analysis that incorporating AI into colonoscopy could potentially reduce the ‘miss rate’ for polyp and adenocarcinoma detection by over 50%.  Similarly, in the mammography setting, he noted a randomized, non-inferiority trial of over 80,000 women having their mammography read using human-assisted AI versus the standard procedure of using two separate radiologists to read questionable scans.  While the outcomes from the trial are pending, Dr. Dicker noted a significant decrease in the overall cognitive burden on the radiologists, with a reduction of nearly 50% in the number of screen readings that were needed.


Cancer Risk Prediction


In the breast cancer realm, Dr. Dicker also highlighted the use of AI, using large and geographically distinct data sets (e.g., US, European, Asian) as well as clinical features, to create very robust prediction models for a patient’s breast cancer risk over five years, based on a single mammogram.  He noted that these models are in the public domain and available to anyone.  In the setting of lung cancer, AI has also been used to predict, from a single low-dose chest CT scan, a person’s risk for developing lung cancer.  Once again, using diverse data sets from tens of thousands of patients to eliminate potential bias, models have been created that accurately predict risk based on only a single CT scan.  Dr. Dicker noted that this can be especially important for high-risk individuals who might not be amenable to or available for regular lung cancer screenings.


Using AI for Eliminating Bias


AI also has the potential to eliminate potential healthcare biases in underserved populations.  For example, in the osteoarthritis setting, it is known that pain scores in underserved patient populations tend to be higher and are not necessarily correlated with the degree of damage based on radiographic imaging.  In one study, an algorithm was developed to correlate radiologic imaging findings of the knee with the degree of pain experienced by the patient.  Notably, the algorithm outperformed the radiologists' findings alone, and Dr. Dicker highlighted the implications of the study, which suggest that the use of AI to generate such algorithms can serve patients better and reduce healthcare disparities in underserved populations. For example, the recommendation for arthroscopies to address knee pain based on AI-assisted methods can be more sensitive than a radiologist’s assessment alone, which could significantly underestimate the need for surgery.


A Growing Potential for ChatGPT and Other AI Technologies


Dr. Dicker related the recent news story of a boy in chronic pain who had been to 17 doctors over three years and was finally correctly diagnosed based on MRI findings using the ChatGPT app.  He noted that ‘large language’ apps such as ChatGPT (GPT=generative, pretrained, transformers) have evolved from scoring about 70% on medical boards (Step 1, Step 2, and Step 3 exams) just a few years ago to now over 90%, a remarkable improvement in performance.  Dr. Dicker described the important concept of “latency,” meaning that, typically, AI algorithms have been trained on medical information for several years. Hence, it is important to grasp that the model may not have arrived at the proposed solution based on the latest medical information.  He noted that more advanced, significant language AI models, such as OpenEvidence, are constantly updated with the latest information.  He also emphasized using AI technologies to simplify important documents for the general public, such as informed consent for participation in clinical trials.  Another beneficial prospect for using AI technologies, Dr. Dicker highlighted, is the generation of text and appropriate verbiage to facilitate insurance approval for critical procedures and potentially reduce financial toxicity for patients.


Lung Cancer: Use of AI in Plasma Proteomics


In the final portion of his presentation, Dr. Dicker briefly described the use of AI to identify biomarkers, measurable markers present in blood that can help determine whether a specific patient will respond to a given type of cancer therapy, such as immunotherapy.  He described an AI-based assay using plasma proteomics or an identifiable protein ‘signature’ of 7,000 and 10,000 individual proteins that can be assessed in a very small blood volume.  The assay can be used to determine, for example, which metastatic lung cancer patients should receive immunotherapy alone versus immunotherapy and chemotherapy.  The model can also predict which patients are more likely to experience severe immune-related adverse events while on immunotherapy and identify possible resistance mechanisms that can arise while receiving treatment.


Summary: Building Deep Empathy with AI


Summarizing his presentation, Dr. Dicker noted that the overall goal of using AI in oncology is not to reduce the time a physician spends with the patient but rather to help the patient make better decisions, decrease some of the enormous burdens on the healthcare providers, reduce costs, and promote greater efficiency in healthcare.  As such, the continued integration of AI methods into oncology can ultimately allow for greater empathy with the patient by better understanding the patient’s cancers, optimizing care pathways, increasing the safety of treatments, and facilitating greater patient engagement with their care.   


Speaker Disclosure Information:  Dr. Dicker is an employee of Thomas Jefferson University/Jefferson Health and reported the following disclosures for this presentation: Current Support: National Cancer Institute; American Society of Clinical Oncology; Prostate Cancer Research Program (Department of Defense); NRG Oncology; Prostate Cancer Foundation; Advisory activities: Active:Janssen, Oranomed, Aptar, SBRBio, Oncohost; Inactive: Wilson Soncini (expert testimony Intellectual Property), Onconova Therapeutics, Bluespark, Imagene, Roche, IBA, CVS; Unpaid advisor: Google LaunchPad Accelerator, Dreamit Ventures, Israel Innovation Authority, University Science Center


You can see Dr Dicker's full presentation from the 2024 West Oncology Conference here:


bottom of page