How Can AI Models Measure Disease Severity from Healthy Data?

Original title: Quantifying Impairment and Disease Severity Using AI Models Trained on Healthy Subjects

Authors: Boyang Yu, Aakash Kaku, Kangning Liu, Avinash Parnandi, Emily Fokas, Anita Venkatesan, Natasha Pandit, Rajesh Ranganath, Heidi Schambra, Carlos Fernandez-Granda

The article introduces a new method, the COBRA score, aiming to assess impairment and disease severity using AI models trained solely on healthy individuals. By detecting deviations in model confidence when analyzing patients with impairments or diseases, COBRA quantifies these deviations from the healthy population. Its application on stroke patients with upper-body impairment, using wearable sensors and video data, displayed a strong correlation with the standard Fugl-Meyer Assessment (FMA). This novel approach offers rapid assessments within a minute, in contrast to the time-consuming FMA, allowing more frequent monitoring and adaptable rehabilitation protocols for each patient. Moreover, the COBRA score proved its versatility by accurately quantifying the severity of knee osteoarthritis from MRI scans, showcasing its potential for various conditions beyond stroke assessment.

Original article: https://arxiv.org/abs/2311.12781