MIT's new AI model detects pancreatic cancer in early stages
Researchers at MIT's CSAIL division have achieved a breakthrough in detecting pancreatic cancer with a "PRISM" neural network, which they have formed using two machine learning algorithms. The PRISM model is said to detect the disease at a higher threshold than current diagnostic standards. It has a 35% detection rate for pancreatic ductal adenocarcinoma (PDAC), the most common form of pancreatic cancer, compared to the standard 10% detection rate.
How was it developed?
Although utilizing AI for diagnostics isn't a novel achievement, MIT's PRISM distinguishes itself through its development process. The model was programmed using over five million electronic health records from US health institutions. It uses routine clinical/lab data, like patient demographics, previous diagnoses, medications, and lab results, to make predictions. Kai Jia, an MIT CSAIL researcher and senior author of the paper, believes this diverse data set is a significant advancement over other PDAC models, usually confined to specific geographic regions.
Addressing late diagnosis of pancreatic cancer
The PRISM project started six years ago with the goal of early PDAC detection, as about 80% of patients are diagnosed too late in cancer development. The AI model predicts the probability of cancer by analyzing electronic health record data along with factors like a patient's age and lifestyle risk factors. However, PRISM's reach is currently limited, as the technology is only available in MIT labs and for select patients in the US. Scaling requires feeding the model diverse datasets.
MIT previously developed AI models for predicting breast cancer risk
This isn't the first time MIT has attempted to create an AI model capable of predicting cancer risk. Previously, it developed a way to train AI models for predicting breast cancer risk using mammogram records, finding that more diverse data sets improve diagnostic accuracy.
AI can alleviate the burden on healthcare professionals
Advancing AI models for predicting cancer likelihood not only enhances patient outcomes through early detection, but also alleviates the burden on overstressed healthcare professionals. The burgeoning potential in the AI diagnostics market has also caught the attention of major tech companies like IBM, which tried to create an AI program to detect breast cancer a year in advance.
