自動化尺度校正與色彩正規化方法用於時序性傷口影像辨識

Abstract

一般傷口如果在 6 至 8 週內沒有癒合現象則稱為慢性傷口。慢性傷口的患者， 相對來說不良於行，回診較麻煩。美國每年平均有六百五十萬人死因為傷口感染 及病發症。在未來的高齡化社會，亟需慢性傷口管理，作為長期照護發展的關鍵 技術。 然而每次拍攝照片時傷口的位置都會有所不同，假如因為拍攝的遠近或角度 而造成傷口的大小失真，將不利於慢性傷口的癒合情形追蹤。因此本論文計畫開 發一個演算法，使得術後的傷口照片可以依據正規色彩卡校正其傷口照片的相對 色彩以及大小，並承接台大醫院-術後傷口影像分析計畫，使得傷口的大小、紅、 腫、發炎等情形能夠判斷得更加精準。 本篇論文著重於傷口影像的校正演算法，以尺度不變的特徵轉換追蹤正規 色彩，並以色彩卡為輔助進行大小調整與色彩正規化。在結果的方面使用了邊界 範圍傷口涵蓋率作為評估方式，校正過的傷口影像擁有 90.25%的傷口涵蓋率，未 校正的影像擁有 88.31%的傷口涵蓋率。除此之外，我們也提出了一個電腦輔助傷 口面積計算的模型，與手動計算的傷口面積相比有 12.42%的偏差。

Chronic wounds are incomplete healing wounds after 6 to 8 weeks. Compared to others, patients with chronic wounds are inconvenient in action and have troubles to return for their check-ups regularly. In USA, there are approximately 6.5 million of patient died of wound infections and complications [1]. In our future aging society, an organized long-term wound care technique is urgently needed. However, the wound image taken by the camera alters due to different angles and different distances among the wounds, which is not conducive of wound tracking. Therefore, in this literature we proposed an image processing algorithm which can calibrate the size and normalize the color of the image according to a color reference card. In our work, we proposed an image calibration algorithm based on a color reference card. Our algorithm evaluates the wound coverage rate based on precision recall method. The calibrated image possesses a wound coverage rate of 90.25% while the original image possesses a coverage rate of 88.31%. Furthermore, we proposed a computer aided wound area measurement prototype with a bias of 12.42% comparing to the area calculated manually.