基于卷积神经网络的眼底图像微动脉瘤检测方法

Detection Method of Retinal Fundus Disease Based on Convolutional Neural Network

糖尿病性视网膜病变是常见的视网膜眼底疾病,随着病情的逐渐加重可引起视觉障碍甚至失明,因此,及早对其诊疗能够抑制病情加重。糖尿病引起的视网膜眼底病变主要包括出现微动脉瘤、渗出物以及眼底出血等情况,微动脉瘤是人工检查糖尿病视网膜病变早期病灶之一,针对糖尿病视网膜病变提出一种基于卷积神经网络的微动脉瘤检测方法,可以辅助眼科医生进行视网膜图像检查,准确快速地筛查出眼底图像是否有微动脉瘤的出现。微动脉瘤的检测方法包括图像预处理、微动脉瘤候选区域提取和微动脉瘤识别。对眼底图像进行灰度修正、图像平滑和图像锐化等操作,可提高图像对比度,降低噪声。使用黑塞矩阵确定微动脉瘤候选区域,然后采用卷积神经网络对病灶部位进行检测,能高效准确地识别微动脉瘤,避免因为人为因素对较早期的病变判断有误,造成病人错过最佳治疗时间。目前我国糖尿病患病人数较多,医疗资源有限,传统依赖于眼科医生主观判断的诊疗方法无法满足社会需求,将人工智能引入医疗图像处理中具有非常重要的意义。

Diabetic retinopathy is a common retinal fundus disease,which can cause visual disturbance and even blindness as the disease gradually worsens. Therefore,early diagnosis and treatment can prevent the disease from getting worse. Retinopathy caused by diabetes mainly includes microaneurysms,exudates,and fundus hemorrhage.Micro-aneurysms are one of the early lesions of diabetic retinopathy in manual examination.This article proposes a micro-aneurysm detection method based on convolutional neural network for diabetic retinopathy,which can assist ophthalmologists to perform retinal image inspection,accurately and quickly screen the fundus image for micro-aneurysms.Micro-aneurysm detection methods include image preprocessing,micro-aneurysm candidate area extraction and micro-aneurysm recognition.In this paper,gray-scale correction,image smoothing and image sharpening are performed on fundus images to improve image contrast and reduce noise. Using morphological methods to determine the candidate regions of micro-aneurysms,and finally using convolutional neural networks to detect the lesions can efficiently and accurately identify microaneurysms. This automatic detection system prevents the patient from missing the best treatment time due to artificial factors in the judgment of the earlier lesions.At present,my country has a large number of diabetic patients and limited medical resources.The traditional diagnosis and treatment methods that rely on the subjective judgment of ophthalmologists cannot meet the current social needs. The introduction of artificial intelligence into medical image processing is of great significance.