Computer-aided Detection of Lesions from Coronary Angiograms Based on Contrast Learning Technique.

Coronary artery disease is one of the prevalent cardiovascular diseases in the world. Clinically, coronary artery angiography (CAG) is the most efficient diagnostic tool for detecting the stenosis caused by the presence of coronary lesions. Here, we proposed a simple but efficient methodology for predicting the coronary arterial block. The technique of classifying the angiograms collected from 369 patients is implemented using the contrast learning approach. ResNet 152 V2 is used as the deep network. Region of interest (ROI) is found for the diseased arteries for deciding the type of treatment procedure. Four different losses were implemented in this two-level classification technique. This framework achieved an accuracy of 0.81 recall of 0.76, precision of 0.86, specificity of 0.87, and F-score of 0.80. A comparative study with the state-of-the-art is carried out to establish the advantage of the proposed method. This computer-aided approach could be implemented by clinicians quite easily. Clinical relevance-An easy, simple, and fast technique of detecting and deciding treatment of coronary artery stenosis was studied using contrast learning and ROI. A computerised model was designed which can draw conclusion from CAGs even in the absence of clinical annotations.