An automatic framework for perioperative risks classification from retinal images of complex congenital heart disease patients

Autor: Wing W. Y. Ng, Honghua Yu, Yunxia Sun, Zhongning Huang, Cong Li, Pingting Zhong, Cankun Zhong, Minghui Xu, Xiaohong Yang, Haicong Liang, Xinran Dong, Qingsheng Peng
Rok vydání: 2021
Předmět:
Zdroj: International Journal of Machine Learning and Cybernetics. 13:471-483
ISSN: 1868-808X
1868-8071
DOI: 10.1007/s13042-021-01419-0
Popis: The number of patient suffering from complex congenital heart diseases (CHDs) increases gradually each year. The perioperative parameters assessment of complex CHDs patients is critical in choosing a suitable surgery method, but there is still a lack of an accurate and interpretable approach to preoperatively assess surgical risks and prognosis. The vascular patterns in retinal images of patients with complex CHDs reflect the severity of heart disease, so retinal images are used to predict the risk of perioperative parameters of heart disease. Perioperative parameters classification from retinal images is challenging due to the limited available retinal image data in patients with CHDs and the interference caused by retinal images with poor quality. In this work, a method called deep learning based perioperative parameter classifier is proposed to classify perioperative parameter risk from retinal images of patients with complex CHDs. To evaluate its effectiveness, our method is verified with 6 perioperative parameters, respectively. Experimental results show that the proposed method is superior to several popular classification networks in this task. Saliency maps are also provided to enhance the interpretability in our model and may be of great use for future medical researches.
Databáze: OpenAIRE
Externí odkaz: