可見光影像游離皮瓣術後監測系統之開發：自動對位與即時追蹤

Abstract

在面對重大事故傷害、病症灶與感染等的大面積複合式組織缺損情況，醫生多會使用游離皮瓣手術為病患進行治療。在過去的研究中顯示游離皮瓣手術的成功率高達91~99%，其中仍有少數案例發生血液循環障礙，因血液循環障礙需再次手術搶救者約有5~25%，因此先前的研究建議手術後2~3天為游離皮瓣手術的重要觀察時段。目前臨床外科醫師廣泛接受的觀察方式為人工監控觀察法，此方法需要高度的仰賴醫護人員的訓練與主觀經驗的判斷，對醫院造成嚴重的人力資源負荷。目前的術後監控輔助設備多數是高耗材成本、高專業程度需求、具侵入性等條件限制。 為了要發展一套符合現今臨床需求具有操作簡便、快速即時、低成本、非侵入接觸、無放射性等性質的監控輔助設備以符臨床所需，本研究團隊研發「紅外線與可見光影像游離皮瓣即時監控系統」，藉由結合漢唐集成的「紅外線熱像儀醫療診斷系統」與一般常用的可見光相機，開發游離皮瓣半自動影像監控之系統，使其具有可即時且連續時間點的自動對位、追蹤、偵測動靜脈阻塞現象、回饋警示血管阻塞狀況等功能之設備。在系統研發的過程中，本研究團隊將其劃分為「持續性紅外光與可見光影像擷取系統」與「游離皮瓣血管阻塞判斷偵測系統」兩部分。本研究內容為「持續性紅外光與可見光影像擷取系統」，主要針對此子系統中可見光影像的部分進行相關研究，目標功能包含皮瓣辨識、自動對位、即時追蹤，取得皮瓣區域的資訊。提供紅外光影像方便取得皮瓣位置，有利於分析皮瓣內的血液循環是否有阻塞之情況。由於皮瓣為位置多樣且大小不一，目前此設計之子系統主要針對頭頸部之游離皮瓣手術血液循環異常之監控。為使系統能夠架設在加護病房中，使用的空間需求較小、容易操作與具有較高的安全性的機械手臂，將紅外光及可見光相機鑲嵌在機械手臂上，使紅外光及可見光相機可同時控制進行拍攝，獲得同時間點及近乎相同視角之影像。本研究開發之子系統為半自動式的監控系統，子系統自動執行前需手動設定皮瓣影像模型與手繪皮瓣邊緣建立皮瓣邊緣影像模型。子系統開始運作時，首先以膚色與臉部比例的方式取得臉部位置，透過尺度不變特徵變換匹配演算法以對位方式與仿射轉換進行人臉影像中皮瓣區域的辨識，在取的皮瓣區域後透過連貫性點群飄移的方式將皮瓣進行對位，接著透過薄板仿樣分析法獲得皮瓣的邊緣座標。 結果中發現，本研究開發的系統已能夠達成連續多時間點的皮瓣偵測、追蹤與自動對位取得皮瓣區域資訊的能力。人臉偵測的結果中顯示，可有效的去除非臉部區域之背景影像，並獲得包含有皮瓣區域的臉部影像。接著將偵測得之臉部區域影像進行皮瓣區域的辨識，尺度不變特徵變換匹配演算法與仿射轉換，將手繪圈選皮瓣區域影像模型之特徵點與偵測得的臉部區域影像之特徵點做特徵點匹配，辨識臉部區域中的皮瓣位置並將此區域經由仿射轉換顯示，獲得皮瓣區域位置。利用皮瓣區域位置中皮瓣影像的紋理和皮瓣邊緣的交叉點以及尺度不變特徵變換匹配演算法的特徵點進行連貫性點群飄移方式的點群對位，使皮瓣影像模型與測得之皮瓣區域影像中的皮瓣對齊，並藉由兩組對位點群計算獲得點群的移動場。最後採用薄板仿樣分析法將手繪皮瓣邊緣影像模型進行形變，使手繪的皮瓣邊緣形變至偵測得皮瓣區域中的皮瓣邊緣上，獲得偵測得皮瓣區域的皮瓣邊緣座標位置，並將偵測得皮瓣區域與皮瓣邊緣設定為新的影像模型，提供為下一張影像進行皮瓣辨識與對位。 本研究的系統已初步達成皮瓣偵測、追蹤與對位的能力，但此系統仍然存有部分的條件限制，例如計算速度須加快、無法對應背景環境快速的變化等，期望未來能夠將此系統更佳的優化，並能夠實際應用在急診室的術後監控之中，提高醫療的品質。

The microvascular free flap surgery is used to reconstruct large-scale areas of complex tissue defects for patients such as major accident injuries, disease foci and infections, etc. Previous studies have showed that the success rate of the microvascular free flap surgery is as high as 91~99%. However, there are still several cases of circulatory compromise. According to statistics, about 5~25% of patients need re-exploration because of circulatory compromise. Therefore, previous studies suggest that 2~3 days after surgery is an important observation period for the microvascular free flap surgery. Manual monitoring observation is currently the most widely used post-surgical monitoring method for clinical surgeons in practice. This method heavily relies on medical personnel with medical training and practical experience, thus causing heavy human resource burden in the hospital. Although several automatic or semi-automatic post-surgical monitoring approaches have been developed, most of them suffer such limitations as high consumables cost, high professional level requirements, invasiveness, and so on, which prevent these methods from practical uses. To develop a real-time, low-cost, non-invasive and non-radioactive monitoring system which can be practically used in Intensive Care Unit (ICU) our team develop the system called “Free-flap Auto-Tracking System (FATS) for Thrombosis of Free Flap after Surgery”. Hardware of FATS was constituted of visible light camera and infrared medical diagnosis system which was made in United Integrated Services (UIS) company. FATS includes the function of automatic detection and tracking simultaneously. Moreover, this system can inform doctors, when detecting the phenomenon of circulatory compromise. FATS is composed of two sub-systems: (I) image acquisition system and (II) free flap analysis system. This study, “Visible Light and Infrared Image Monitoring System for Thrombosis of Free Flap after Surgery: Auto-registration and Real-time Tracking of Free Flap”, is the (I) part of FATS, and with feature of flap detection, flap alignment and tracking instantaneously. The infrared image of the flap position can be used to analyze whether the circulatory compromise happened in the flap or not. Due to a variety of positions and the different sizes of the flaps, the system is aimed at monitoring the circulatory compromise of the free flap surgery of the head and neck. In order to mount the infrared light and visible light cameras on the robot arm and be applied in ICU, this system needs to possess the characteristics of small space requirements, easy operation and high safety. Firstly, the position of the face is obtained by the skin color and the proportion of the face. Secondly, the scale-invariant feature transformation (SIFT) algorithm and affine transformation were used to identify flap regions. Finally, the edge of the flap is aligned by the coherent point drift (CPD) method, and the edge coordinates of the flap are obtained by the thin-plate spline (TPS) method. The result of this study shows that the system has been able to achieve flap detection, tracking and automatic alignment of the flap area information for multiple time points. Obtaining a face area with the flap by face detection. Then, the flap region is detected by SIFT alignment from the face-detected area. Furthermore, we matched the feature points of the flap region selected by the hand-drawn circle with the detected face area feature points to obtain the position of the flap region of face-detected area. To align the model of hand-drawn of the flap and face-detected area, cross point of the flap and SIFT feature point were used in CPD method. Additionally, we can obtain the moving field from the point group after CPD. Lastly, the TPS method can transform the hand-drawn flap edge on the face-detected flap region to obtain the flap edge position. Although the system of this study possesses the merits of flap detection, tracking and alignment, it still has some conditional restrictions, including that the speed of calculation must be accelerated, and it is unable to respond to rapid changes in the background. It is expected that this system will be better optimized in the future, and can be practically applied in the postoperative monitoring of ICU to improve the quality of medical care.