MIT's Artificial Intelligence Laboratory (MIT CSAIL), in collaboration with Dr. Limor Apelbaum from Harvard University's Department of Radiation Oncology, has developed two new AI-based models, PRISM Neural Network (PrismNN) and Logistic Regression (PrismLR), for early detection of pancreatic ductal adenocarcinoma (PDAC), a deadly type of cancer. Why is this discovery important? The results show that PRISM can identify 35% of PDAC cases in situations with relatively high risk, compared to only 10% identified by traditional screening criteria. This improvement in performance signifies a significant breakthrough in early intervention potential. The location of the pancreas deep within the abdomen makes early detection challenging, and the lack of effective treatment methods further emphasizes the importance of identifying high-risk patients early. Therefore, the research team leveraged the advantages of alliance network companies and utilized electronic health records (EHR) data from various institutions in the United States, including patient demographics, diagnoses, medications, and laboratory results. This extensive database covers over five million patients, ensuring the reliability and applicability of the models across different populations, geographical locations, and demographics. PrismNN employs artificial neural networks to detect complex patterns, while PrismLR uses logistic regression for simpler analysis, providing a comprehensive assessment of PDAC risk from the same EHR data. Although the use of AI in cancer risk detection is not new, what sets PRISM apart is that it was developed and validated on a database of over five million patients, surpassing the scale of most previous studies in this field. "The model uses conventional clinical and laboratory data for prediction, and its diversity in data is a significant advancement compared to other PDAC models, which are often limited to specific geographic regions, such as some medical centers in the United States. Additionally, the model's generalizability and interpretability are enhanced through the use of unique regularization techniques during the training process," he added. The team envisions expanding the model's applicability to international datasets beyond the United States and incorporating more biomarkers for a more refined risk assessment. Previously, Google DeepMind's protein folding AI system, AlphaFold, helped accelerate the design and synthesis of drugs for the treatment of hepatocellular carcinoma (HCC), the most common type of primary liver cancer.