Risk Rating of Infantile Hemangioma using Deep Learning

Abstract

Infantile hemangioma is one of the most common benign tumors, which appears in the early stages of life, most of which can be cured automatically, but some serious cases can threaten the normal growth and even life of the baby. Therefore, making timely and correct risk ratings for the status of hemangioma is extremely important for the treatment of patients. At present, this work is mainly done manually by pediatricians with high professional quality. This study proposes a deep learning-based method to rank infant hemangioma risk, which is divided into three levels: high risk, medium risk and low risk. This article describes a hemangioma risk classifier based on a convolutional neural network structure to achieve an assessment of the risk of hemangioma for auxiliary diagnosis. The challenge is how to achieve good classification on a relatively small data set, which contains 1032 images from 344 different patients. The final result is promising, according to the performance evaluation of the model, the accuracy on the test set reaches 90.85%.