Ethical Multimodal fusion Artificial Intelligence platform (EMAP) to bridge the gap between imaging modalities: Application to prostate cancer diagnosis.

Stanford University, Stanford CA

Investigators

Abstract

ABSTRACT Ultrasound guides nearly all prostate biopsies. However, low contrast between cancer and normal tissue on ultrasound limits its utilization in identifying suspicious regions to target. Instead, ultrasound-only biopsies blindly sample the prostate at ∼12-14 locations, missing 52% of cancers. When combined with Magnetic Resonance Imaging (MRI), the Ultrasound-MRI fusion biopsy enables targeting of suspicious lesions for better cancer detection (88% vs. 48%). Yet, MRI is only available in 35% of biopsies due to cost and the need for interpretation expertise. Unequal access to MRI represents a major health disparity. Despite being at higher risk for prostate cancer, Black men are 50% less likely to get an MRI, leaving them to undergo sub-optimal blind biopsies, possibly delaying diagnosis. There is a clinical need to enable prostate cancer targeting in low-resource settings, using only ultrasound when MRI is inaccessible, and improve targeting in resource-rich settings using both ultrasound and MRI. We propose to develop the Ethical Multimodal AI fusion Platform (eMAP) and show proof-of-concept for prostate cancer detection. eMAP is built under careful ethical considerations (e.g, data privacy, fairness, stakeholder input) using advanced AI foundation models and knowledge transfer architectures. eMAP integrates multimodal AI training allowing MRI to inform ultrasound to enhance cancer features, but at inference, in unseen subjects, eMAP will allow improved cancer detection in both resource low and rich settings. While we show proof-of-concept in prostate cancer, we anticipate that eMAP will pave the way to ethically bridge the gap between radiology imaging modalities in prostate and beyond.