New Method Advances AI Reliability with Applications in Medical Diagnostics

Circulating cell-free DNA (ccfDNA) fragmentation patterns, once believed to be cancer-specific, were recently found to also occur in autoimmune and vascular diseases. This overlap complicates the development of accurate diagnostic tests, as inflammation can trigger signals easily misconstrued as cancer. Now, researchers have developed a new artificial intelligence (AI)-driven method to improve sensitivity while reducing false positives.

The innovation, called MIGHT (Multidimensional Informed Generalized Hypothesis Testing), was created by a research team at Johns Hopkins Medicine (Baltimore, MD, USA) to help achieve the high level of confidence required for clinical decision-making. MIGHT uses tens of thousands of decision trees to measure uncertainty and fine-tune its predictions. It can be applied across any field with big data, but is especially effective for biomedical datasets with many variables and relatively few patient samples.

MIGHT works by evaluating multiple sets of biological features in blood samples, such as DNA fragment lengths or chromosomal abnormalities. It was further extended to a companion tool, CoMIGHT, which can combine variable sets to improve detection. By incorporating inflammation-related data into training, the algorithm distinguishes more accurately between cancer-related signals and those caused by other diseases, thereby reducing false-positive results.

The studies, published in the Proceedings of the National Academy of Sciences and Cancer Discovery, showed promising results. In a trial of 1,000 participants, including 352 cancer patients, MIGHT achieved 72% sensitivity and 98% specificity using aneuploidy-based features. CoMIGHT, applied to 125 patients with breast cancer, 125 with pancreatic cancer, and 500 controls, revealed that combining biological signals improved early breast cancer detection, while pancreatic cancers were more readily identified.

These findings underscore the complexities of AI-informed diagnostics but also highlight significant potential. By addressing the problem of false positives caused by inflammation, MIGHT not only advances cancer detection but may also pave the way for new diagnostic tests targeting autoimmune and vascular diseases. Researchers stress that further clinical trials are needed before clinical rollout, but note that MIGHT and CoMIGHT are now publicly available for wider testing through treeple.ai.

“Trust in the result is essential, and now that there is a reliable, quantitative tool in MIGHT, we and other researchers can use it and focus our efforts on studying more patients and adding statistically meaningful features to our tests for earlier cancer detection,” said Bert Vogelstein, M.D., study co-leader.

Related Links:
Johns Hopkins Medicine