In current clinical practice, assessment of impairment and disease severity typically relies on examinations by medical professionals. As a result, assessment is often qualitative and its frequency is constrained by clinician availability. Developing data-driven quantitative metrics of impairment and disease severity has the potential to enable continuous and objective monitoring of patient recovery or decline. Such monitoring would facilitate personalized treatment and administration of appropriate therapeutic interventions in telehealth and other remotely supervised contexts where ongoing access to clinicians is not readily available

In this work we present a method to perform automatic assessment of impairment and disease severity using AI models trained only on healthy individuals. The COnfidence-Based chaRacterization of Anomalies (COBRA) score exploits the decrease in confidence of these models when processing data from impaired or diseased patients to quantify their deviation from the healthy population.

The following diagram illustrates the method, which consists of two steps. In Step 1, an AI model is trained to perform a clinically meaningful task on data from healthy individuals. For impairment quantification in stroke patients, the task is prediction of functional motions from videos or wearable-sensor data (top). For severity quantification of knee osteoarthritis, the task is segmentation of knee tissues from magnetic resonance imaging scans (bottom). In Step 2, the COBRA score is computed based on the confidence of the AI model when performing the task on patient data. Data from patients with higher degrees of impairment or severity differ more from the healthy population used for training, which results in decreased model confidence and hence a lower COBRA score.

Our results show that the COBRA score is highly correlated with expert-based metrics for two medical conditions (stroke and knee osteoarthritis), and three different data  modalities (wearable sensors, videos and magnetic-resonance scans). This suggests that confidence-based anomaly detection is a promising methodology to build data-driven metrics for the quantification of impairment and disease severity.  See the article or this YouTube Video for more details.