Intelligent Diagnosis of Vascular Anomalies with Deep Learning

Abstract

Correct classification and diagnosis of vascular anomalies is an important prerequisite for further clinical treatment. In general, clinical diagnosis of vascular anomalies is conducted according to different characteristics of ultrasonography. However, the clinical presentation of infantile hemangioma (IH) and venous mal- formation (VM) is similar and not easy to be separated. The difficulties result from the special diagnostic ultrasonography mode, indistinguishable grayscale images, rich texture, extensive involvement area, blurred boundary, and the unique blood flow ultrasonography acquisition method of VM. Here we propose a deep learning algorithm for detection and recognition of IH and VM lesions from ultrasonography. The experimental results have shown a good recognition performance on IH and VM ultrasonic images.