使用圖形處理器加速糖尿病視網膜病變偵測

Abstract

根據世界衛生組織統計，預估2030年糖尿病患者將會成長至366百萬人，其中因糖尿病而引發的糖尿病視網膜病變是成年人致盲主因之一，其病變特徵有微血管瘤、出血、滲出物等。本研究發展出一套糖尿病視網膜病變眼底影像量化評估系統，擷取視網膜病變特徵，並以支援向量機進行病變的分類與識別，可以辨識並計算病變的區域面積與成長趨勢等量化資訊，其正確率可達94％。

According to the World Health Organization, the total number of diabetic patients will grow to 366 million in 2030. One of serious complications caused by diabetes is retinopathy which will lead to blindness. The symptoms of retinopathy include microaneurysms, hemorrhages, and exudates. This paper proposes a quantitative evaluation system for diabetic retinopathy fundus image. The system extracts retinal lesion features and uses support vector machine for lesion classification. The system achieves an average of 94% of accuracy for lesion identification.