Abstract

The following whitepaper details the operating characteristics and bench testing performance of the algorithms within AI-Rad Companion Chest X-ray VA23A and newer versions. It describes the general principles of the software such as its workflow and results. The white-paper also explains performance metrics like area under receiver operating characteristics curve, sensitivity, specificity etc. and how to interpret these algorithm performance metrics. It also delves into how the device performance can be benchmarked in comparison to average performance of board-certified radiologists. Towards this, a multi-case multi-reader study has been conducted to measure the average reader performance of radiologists for analysis of chest X-rays and compare it with the standalone performance of AI-Rad Companion Chest X-ray VA23A. The target radiographic findings include pulmonary lesions, atelectasis, pneumothorax, consolidation, and pleural effusion. The conclusions of the study are used as a measurable baseline to quantify standard-of-care performance. It has been demonstrated that AI-Rad Companion Chest X-ray VA23A performs comparable to the average performance of radiologists participating in the study with higher area under receiver operating characteristics curve and sensitivity for all the target radiographic findings. With AUC of 95–99 %, the AI algorithms within AI-Rad Companion Chest X-ray demonstrate high accuracy for the detection of the target radiographic findings for use as a diagnostic aid for concurrent reading of Chest X-rays.

Click the below image to read the full document: