I-Corps: Translation potential of an artificial intelligence (AI) approach for brain lesion segmentation

This I-Corps project is based on the development of a generative artificial intelligence (AI)-driven platform to enhance diagnostic accuracy in medical imaging. This solution may accelerate the diagnostic and treatment processes. In addition, the platform uses explainability, which supports early detection, diagnosis, triage prioritization, and treatment planning, particularly for surgeries.

The technology is designed to enhance medical practices in imaging centers and hospitals. It may be used to sharpen the precision of surgeries on brain tumors, reducing operation times and fostering quicker patient recovery. This technology may lead to lower healthcare costs and a reduced need for repeat surgeries, ultimately boosting the overall quality of patient care and outcomes.

In addition, the technology may be used in emergency rooms to differentiate between types of strokes —conditions that require distinctly different treatments. This ability to distinguish types of strokes may improve the speed and accuracy of stroke diagnosis, helping patients receive the right treatment at the right time, enhancing outcomes in critical care situations.

This I-Corps project utilizes experiential learning coupled with a first-hand investigation of the industry ecosystem to assess the translation potential of neuroimaging algorithms for improving brain tumor segmentation and stroke diagnosis. The technology is designed to identify the boundaries of various brain malformations to enhance diagnostic accuracy in medical imaging.

The platform employs explainability, which helps physicians understand the algorithm's suggestions, and ultimately enhances decision-making. This technology leverages the power of advanced, trustworthy neural network models designed for high-precision brain tumor segmentation as well as a deep-learning tool for stroke lesion identification and classification.

This advancement may improve the accuracy of tumor segmentation over current tools. The technology is based on the development of an end-to-end object detection model. The model uses a vision transformer as its backbone to effectively identify and segment the stroke lesion areas.

This tool also employs adaptive learning techniques that continuously refine its performance. Research findings show that the tool recognizes the complex variations in tumor shapes and sizes. This ability may allow the technology to adapt and evolve, providing reliable and enhanced diagnostic capabilities and improving patient outcomes.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.