Annual Report 2012–13

Pushing Knowledge Frontiers 25 偵測眼內增生的血管是最大的難題，它們 一旦出現，就是大禍逼近的先兆。增生的血 管既短且彎，形狀不規則，要發現並查出它 們的生長情況並不容易。團隊使出從雜亂 無章的資料中識別模式的工夫，設計了一套 運算法，把視網膜圖像分拆為像素單位，逐 點分析，以量度眼底滲液、出血和血管增生 情況，從而得出視網膜病變的整體評估結 果。 一般視網膜圖像可以透過互聯網，傳送至 內置演算法的伺服器，不一會即可取得檢 查報告。這種嶄新的非入侵檢查方法，不受 體檢人員的主觀影響，省錢省時。初步測試 證實可靠，而且準確度高。下一步，徐教授 將會把技術應用到糖尿及非糖尿病人身上， 及早防範中風。 To address these issues, the team had to find a methodology to turn the analogue images of the retina into quantifiable and analyzable data. The first difficulty encountered in the process was the location of the optic disc (the spot where the light-sensitive ganglion cell axons leave the eye to form the optic nerve to the brain, also known as the physiological blind spot). Although methods of locating it already existed, his team developed a new method that best fit their purpose. A greater hurdle laid in the detection of new vessels in the eye whose growth is a sure sign of havoc to come. The new blood vessels are short, irregular and squiggly. Applying pattern recognition skills, the team was able to devise an algorithm which reads pixel by pixel the retinal images and analyze such pixels to come up with measurements on exudates, haemorrhages, new vessels and finally achieve the overall evaluation of retinopathy. Standard retinal images can be transmitted through the Internet to a server installed with the algorithm and the result is available within a short period of time. The new method is non-invasive and will substantially reduce bias due to human perception as well as cost and time. Initial tests have confirmed its dependability and high accuracy rates. Next, Professor Zee intends to expand the technology and apply it to both diabetes and non-diabetes patients for the early detection of strokes. 左起：徐仲鍈教授、李作為博士、莫仲棠教授 From left: Prof. Benny Zee, Dr. Jack Lee, Prof. Vincent Mok