Coronary Plaque Classification of Intravascular Ultrasound Images based on a Multi-Stage Deep Classifier Cascades

A method of multi-stage deep classifier cascades is proposed to classify coronary plaque from gray-scale IVUS images. The framework consists of one segmentation module and four classification modules, which are trained and evaluated successively. The pre-segmentation module aims to highlight potential plaque territory. A multi-level classification strategy is designed based on plaque histological composition and morphology. The IVUS images are labelled by clinical cardiologists and used to evaluate the effectiveness of multi-stage classifier. The result shows that proposed framework improves the accuracy of coronary plaque classification.