複雜性軟組織感染性疾病紅外線熱影像療效追蹤系統:基於三維表面多時間點對位演算法

Abstract

皮膚軟組織感染是一常見的感染性疾病。根據美國感染症醫學會制定之治療指引，臨床診斷主要依賴患處之身體檢查：在觸診時以雙手觸摸患部與其鏡像解剖構造之對側位置，比較其溫度之差異程度，加上白血球計數、C-反應蛋白等檢驗，及核磁共振MRI影像作為參考依據。然而，臨床上之身體檢查中，熱或紅等理學檢查結果之判讀標準大多難以量化;臨床指標如發炎指數、白血球數雖可量化，但治療後此些指數皆可能快速下降，難以作為終止治療單一的指標；而影像學方面MRI雖為建議之參考標準，但其價格昂貴且具有腎毒性之風險，不適合用以密集追蹤治療療效。因此本研究提出以紅外線熱影像追縱複雜性軟組織感染的評估療效，藉由觀察紅外線熱影像及相同感興趣區域於治療過程中之熱變化以評估療效。 紅外線為一非侵入性、可攜式且可將熱量化之工具。以紅外線熱影像追蹤之關鍵，在於如何比較多次追蹤中不同時間點所拍攝之紅外線熱影像，因此需克服角度、姿勢等造成的形變問題。本研究提出利用三維掃描表面作為媒介，間接地將多時間點二維多時間點的紅外線熱影像對位，其中透過2D/3D紅外光影像與掃描表面對位演算法，取得具有溫度分布的三維掃描表面，再以CPD 3D掃描表面間之對位演算法，間接地完成不同時間點紅外線熱影像間的對位，並提出不同時間點紅外線熱影像之正規化方法，使紅外線熱影像在相似基礎下進行觀察與分析。 透過本研究之結果顯示，有效完成之多時間點紅外線熱影像之對位，可克服因拍攝紅外線之角度以及距離，紅外線熱影像之正規化方面，能夠使分析病患的紅外線影像中的基礎溫度相似，因此本研究不但能夠藉由多時間點紅外線熱影像對位結果清楚地呈現溫度增高之區域，也能夠有效地分析相同區域之溫度變化。

Skin and soft tissue infection (SSTI) is a major infectious disease. According to the latest Infectious Disease Society of America (IDSA) guideline for SSTI, the diagnosis is mainly based on physical examination of infected site with the local heat, redness, swelling, and pain. The laboratory tests can include white blood cell count (WBC) and C-reactive protein (CRP). The Magnetic Resonance Imaging (MRI) will be considered if deep site infection is suspected as a diagnostic or follow-up tool. However, the longitudinal follow up of treatment outcome for SSTI is usually difficult. The local infection signs are subjective and hard to be quantified. Even though WBC and CRP can be quantified, they are usually returned to normal range after initial antibiotics treatment and can’t be relied as a single marker to determine treatment duration. For MRI, the evaluation tool suggested by guideline, it may not be available for every hospitals, costly, and still carries the risk for nephrotoxicity. Thus, it may not be suitable as a continuous monitoring tool. Thus, we proposed an infrared thermometer system, non-invasive, portable and heat-generating tool, to follow up of the treatment response for SSTI. The key is how to analyze the local temperature in the same region of interest (ROI) among infrared thermal images taken at different time, which should overcome the deformation problems caused by different angle and posture because there is no available landmarks for longitudinal registration. To overcome the above possible technique gaps, we proposed 3D scanning surface as the transformation media to register the longitudinal infrared image. The main concept was to obtain the 3D scanning surface with temperature at first and then registered them each other. In order to gain more accurate registration results, the visible image with higher resolution was initially registered with the 3D scanning surface by camera calibration at first, and then calculated the correspondence between the visible image and infrared thermal image by the homography matrix coordinate transformation algorithm. Finally, the registration of 3D scanning surfaces used a point set approach, Coherent Point Drift (CPD), extracted by Growing Neural Gas algorithm. To avoid being affected by the non-fixed markers in cross-sectional images, the feature point sets on it would be removed. We also proposed the normalization method for infrared thermal images based on the registered result that could make the basic temperature similar in cross-sectional infrared images. Therefore, this study could not only clearly show the temperature change through the infrared thermal image registration result, but also effectively analyze temperature changes in infrared images.