可学习的硬性渗出病变点标注方法

Learnable Method for Labeling Hard Exudates

眼底图像的定期筛查对许多疾病的早期诊断治疗具有重要意义,但患者数量的急剧增加对医护人员造成了巨大的压力,阅片工作占用了眼科医生大量的时间,疲劳审阅、情绪等因素都会影响医生的诊断结果.计算机辅助诊断工具的出现可以缓解医生阅片的压力,提高医生的诊断效率和准确率.视网膜辅助诊断工具的研究需要大量精细标注的眼底图像,在眼底数据集的图片数量无法满足研究需求时,本文提出了一种具有学习能力的硬性渗出病变点标注方法,使用形态学和机器学习的方法分割硬性渗出,将分割结果作为用户的预标注,用户只需简单点击校对标注结果,就能够在短时间内完成像素级的标注,大大提高了眼底图像的标注速度.

Regular screening of fundus images is important for the early diagnosis and treatment of many diseases,but the dramatic increase in the number of patients has put tremendous pressure on medical staff.The reading work takes up a lot of time from the ophthalmologist.Factors such as fatigue and mood can affect the doctor’s diagnosis.The emergence of computer-aided diagnostic tools can alleviate the pressure on doctors to read and improve the diagnostic efficiency and accuracy of doctors.The research of retinal-assisted diagnostic tools requires a large number of fine-labeled fundus images.When the number of fundus data sets cannot meet the research needs,this paper proposes a learnable method for hardexudates,which uses the morphology and machine learning methods to segment the exudate.The segmentation result is pre-labeled for the user,and the user can complete the pixel-by-pixel labeling in a short time by simply clicking the proofreading result,which greatly improves the labeling speed of the fundus image.