利用手術器械的追蹤與動向分析達成腹腔鏡手術自動化評估系統

Abstract

腹腔鏡手術有許多種訓練方式，最常被使用的是訓練箱，箱內放置假器官，箱上有兩小孔可放入攝影機和手術器械。訓練流程就是先讓學員操作一次模擬手術，再由評估者對錄下的影像做評估，此方法在學員人數稍多的情況下，將非常耗時，學員不能馬上得到評估結果，雖然可以靠增加評估人員去減少等候的時間，但每位評估人員各有主觀標準，也容易讓學員產生矛盾。如果能將評估的過程用一自動化的系統去取代，那麼不僅評估的標準統一，同時也增進評估的效率，學員越快得到回饋的結果也越快能進行技巧的改進。這篇論文希望研究出一個手術器械的追蹤與動向分析的方法並降低錯誤率，最後以此探討發展出自動化評估系統的可能性。 在追蹤器械方面我們利用器械在畫面是長長一條的特性，用霍夫變換(Hough transform)找出畫面中的直線，再將找到的直線兩兩配對，並且為每個配對標記頭尾，我們假定在所有配對中，正確配對的兩直線夾住的區域即為器械本身，利用前後張差距不大的關係以高斯機率密度函數(Gaussian probability density function)將每張畫面最可能的配對挑出，再利用挑出的頭尾畫出軌跡，並由軌跡分析出此操作者的熟練程度。

There are many way to train laparoscopic surgery skill, the most used one would be using training box, with some fake organs inside and two little holes on top for camera and surgery equipment. The training procedure will be letting trainees do one simulate surgery first then evaluator evaluates how well the trainee performs according the video tape recorded during trainees’ simulate surgery procedure. But it will take much times with trainee’s number grows, every trainee cannot get their feedback right away after they finish the simulate surgery, although we can reduce the time waiting by increasing the number of evaluator. But every evaluator has their subjective criterion which will make contradictory if the trainee receives two feedbacks with two totally different suggestions to improve his skill. So if the evaluating procedure can be replaced by an automated system, then not only we can unify the criterion but also reducing the time waiting for feedback. The faster the trainee get their feedback, the faster the trainee improve their skills, the purpose of this paper would be studying a way to do tracking and motion analysis of surgical tools, and reduce the error rate for bad tracking, finally investigates the probability to build automated system. We use the fact that equipment used in operation is always to show a long stick in screen. So the Hough transform is used to find straight lines in screen and put every line we find into pairs of two and then mark these pairs the head and the tail. We assume among these pairs, the right pair would include the area of equipment stick. Because of the difference between frame and frame is usually small, the Bayesian classification will find the most probable pair in each frame. So the heads of chosen pairs will form the trajectory of the operation equipment, according the trajectory the system then can give a performance evaluation of how well the operator performs.