Search This Blog

Powered by Blogger.

Blog Archive

Labels

Showing posts with label AI-based risk prediction. Show all posts

New AI System Aids Early Detection of Deadly Pancreatic Cancer Cases

 

A new research has unveiled a novel AI system designed to enhance the detection of the most prevalent type of pancreatic cancer. Identifying pancreatic cancer poses challenges due to the pancreas being obscured by surrounding organs, making tumor identification challenging. Moreover, symptoms rarely manifest in early stages, resulting in diagnoses at advanced stages when the cancer has already spread, diminishing chances of a cure.

To address this, a collaborative effort between MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Limor Appelbaum from Beth Israel Deaconess Medical Center produced an AI system aimed at predicting the likelihood of an individual developing pancreatic ductal adenocarcinoma (PDAC), the predominant form of the cancer. This AI system, named PRISM, demonstrated superior performance compared to existing diagnostic standards, presenting the potential for future clinical applications in identifying candidates for early screening or testing, ultimately leading to improved outcomes.

The researchers aspired to construct a model capable of forecasting a patient's risk of PDAC diagnosis within the next six to 18 months, facilitating early detection and treatment. Leveraging existing electronic health records, the PRISM system comprises two AI models. The first model, utilizing artificial neural networks, analyzes patterns in data such as age, medical history, and lab results to calculate a personalized risk score. The second model, employing a simpler algorithm, processes the same data to generate a comparable score.

The team fed anonymized data from 6 million electronic health records, including 35,387 PDAC cases, from 55 U.S. healthcare organizations into the models. By evaluating PDAC risk every 90 days, the neural network identified 35% of eventual pancreatic cancer cases as high risk six to 18 months before diagnosis, signifying a notable advancement over existing screening systems. With pancreatic cancer lacking routine screening recommendations for the general population, the current criteria capture only around 10% of cases.

While the AI system shows promise in early detection, experts caution that the model's impact depends on its ability to identify cases early enough for effective treatment. Michael Goggins, a pancreatic cancer specialist at Johns Hopkins University School of Medicine, emphasizes the importance of early detection and acknowledges the potential improvement offered by the PRISM system.

The study, while retrospective, sets the groundwork for future investigations involving real-time data and outcome assessments. The research team acknowledges potential challenges related to the generalizability of AI models across different healthcare organizations, emphasizing the need for diverse datasets. PRISM holds promise for deployment in two ways: selectively recommending pancreatic cancer testing for specific patients and initiating broader screenings using blood or saliva tests for asymptomatic individuals. Limor Appelbaum envisions the transition of such models from academic literature to clinical practice, emphasizing their life-saving potential.