紅外線影像游離皮瓣術後監測系統之開發：自動對位與溫度變化偵測

Abstract

口腔癌(Oral Cancer)是目前癌症中，發生率與死亡率成長速度最快的癌症；在口腔癌的治療方面，透過外科手術切除病灶，是目前臨床中最常使用的方法，當病患切除口腔癌病灶後，需要利用病患身體上其他部位的組織進行患處的填補，其中，游離皮瓣手術是多年來可靠又重要重建方法。然而，儘管由最有經驗的顯微手術醫師執行游離皮瓣重建手術，仍有少數游離皮瓣會在五至七天內發生血循障礙，且自得知游離皮瓣阻塞到執行再探查的間隔時間越長，也會增加手術的失敗機率，因此如何監控游離皮瓣血液灌流狀況，提早偵測發生血管阻塞時間點，並且進一步即時警示醫師執行再探查，或重新手術接合，非常至關重要！ 目前臨床游離皮瓣照護與監控上，主要是透過護理人員定期進行游離皮瓣的常規檢查，這不但需要非常大的人力資源，也會因為護理人員照護經驗的不同而有判斷的差異；其他監控游離皮瓣血管是否阻塞的方法，不但需要較高的成本，也不能提供一套非侵入式與非接觸式，甚至是連續的監控系統；因此，在先期研究中，提出利用紅外線具備低成本、非侵入性、非接觸性、無放射性、快速且可反覆成像的等特性，在不接觸游離皮瓣表面的情況下，測量其組織溫度的變化，是一項重要的起步；先期研究中，透過手動選取特徵的方式完成影像對位，並且結合因素分析與特徵值分析的方法，完成初步的游離皮瓣手術後血管阻塞之紅外線監控技術。 但透過手動的方式進行影像對位，非常仰賴經驗與耗費時間，無法及時偵測游離皮瓣溫度的變化，且在因素分析演算法有些許誤差，無法精確的觀察皮瓣表面溫度的變化趨勢；因此本研究基於先期研究的技術，發展一套紅外線影像游離皮瓣術後監測系統，其中包含多時間點紅外線熱影像自動對位演算法以及溫度變化偵測演算法兩個部分。 在本研究發展之多時間點紅外線熱影像自動對位演算法中，利用Homography Matrix座標轉換演算法，將可見光影像上偵測之游離皮瓣邊緣特徵點轉換至紅外線熱影像上，以解決先期研究中，需要手動選取紅外線熱影像皮瓣邊緣特徵點繁雜的步驟；獲得紅外線熱影像特徵點後，利用Non-rigid的Coherence Point Drift(CPD)計算多時間點影像特徵點的對應關係，並利用此對應關係，透過仿射轉換(Affine Transformation)來進行游離皮瓣的對位，使同一時間序列上之紅外線熱影像游離皮瓣區域對應在一起，完成自動對位。在完成對位後，本研究利用校正過後的因素分析演算法(Factor Analysis)之方法，分析病患游離皮瓣溫度的變化，並且觀察皮瓣是否發生阻塞的現象。 目前本研究於臨床收案之病患總共10位，其中有一位病患發生靜脈阻塞的情況，經因素分析紅外線熱影像溫度變化的結果，發現游離皮瓣阻塞之皮瓣溫度有明顯下降的現象，而此現象與護理紀錄比對顯示，其游離皮瓣溫度變化有早於護理人員之判斷的可能性，但由於目前阻塞案例太少，因此無法做更進一步的驗證與討論，希望在未來能夠有更多資料進行皮瓣阻塞溫度變化的驗證，並發展出一套提供臨床監測皮瓣血管阻塞的輔助監測工具。

Oral cancer is the fastest growing cancer in mortality rate in Taiwan. The most prevalent curative treatment of oral cancer exploits a multi-step approach starting with the resection of lesions. After the removal of cancer lesions, surgeons would employ unaffected tissues of the patients to reconstruct the affected regions. Free flap surgery is a reliable reconstruction method operated by many medical professionals. However, circulatory compromises were sometimes observed within five to seven days of surgeries, even when the operations were performed by experienced microvascular surgeons. Furthermore, it is documented that the success rate of re-surgeries diminishes as the detection of circulatory compromises delays. Therefore, it is crucial to monitor and detect signs of circulatory compromises quickly after the free flap surgery. At present, clinical free flap care and monitoring are performed mainly through scheduled inspections by nursing staff. This process is very labor-intensive and the success rate of detection varies greatly based on the opinions and experiences of the caregivers. Other currently available methods of monitoring, including the use of non-continuous monitoring systems, are invasive and incur higher costs. It is evident that an alternative method of monitoring is imminent. Previous scholars proposed infrared imaging as an inexpensive, non-invasive, non-contact, non-radioactive, rapid, and repeatable alternative to current monitoring methods. It was proposed that the image registration can be accomplished by manually selecting features points and utilize both factor analysis and eigenvalue analysis to observe temperature variation. However, such process was hand-operated and relies heavily on technician experience and time-consuming. Detection of temperature changes in free flap was prone to errors in the factor analysis algorithm. This study aimed to resolve the disadvantages of the manual image registration process by introducing an automatic longitudinal infrared image registration algorithm and a flap temperature variation detection algorithm. In this study, automatic longitudinal infrared image registration algorithm, utilizing the homography matrix coordinate transformation algorithm, was used to convert the free flap edge feature points detected under the visible light to infrared thermal images. After obtaining the infrared thermal image feature points, this study utilized non-rigid Coherence Point Drift (CPD) to calculate the corresponding relationship of the feature points. Using this relationship, the image registration was completed by affine transformation, so that the free flap regions at different time series were mapped together. After the registration, factor analysis was employed to analyze the temperature variation of free flaps to observe thrombosis. Out of the ten clinical cases monitored by this study, one patient developed venous thrombosis. Analysis using the methods proposed by this study indicated a significant temperature decrease in the subject's free flap. In comparison to nursing records, it was discovered that the methods proposed by this study had the potential to detect temperature changes earlier. However, due to the sample size limitation, this study was unable to provide further verification and discussion. It is hoped that more data will be available in the future to support the proposed analytical methods. The goal of this study it to setup a foundation in the development of an auxiliary monitoring tool to monitor free flap pedicle thrombosis after a lesion removal surgery.