The FDA-cleared ICH-detection AI solution, based on a convolutional neural network, analyzed all technically adequate CT brain exams performed at Alfred Health between Oct. 2019 and Nov. 2019. An independent radiologist retrospectively second-read all CT scans and reports in this timeframe to confirm ICH presence. The independent radiologist then reviewed the output of the AI solution for True/False-Positives/Negatives.
1,269 CT brain scans were analyzed by the AI solution. The independent radiologist confirmed that 184 were ICH-positive and 1,085 were ICH-negative. The sensitivity and specificity of the AI solution were 90.8% (n=167 TP) and 97.2% (n=1,055 TN), respectively. The PPV and NPV were 84.8% and 98.4%, respectively.
The AI solution has high sensitivity, specificity and PPV and helps drives the prioritization of positive cases in turn leading to increased efficiency and reliability. Increased efficiency in prioritization of positive cases has shown to improve patient outcomes.