放射線治療是目前多種癌症治療方式的重要選項,由於高能的放射線必須精準的照射到病灶區,否則周圍鄰近組織將因暴露在治療照野中而引發副作用。因此在治療過程中能夠偵測位移差異並適時提出警告相形重要。本文提出ScanNet位移偵測方式,試圖在每次治療過程中發揮即時偵測病患本身的呼吸、咳嗽或其他原因所造成治療部位的移位,當位移量超過醫療所容許之範圍則發出警告訊息以供醫療人員判斷是否終止醫療行為。 在過去幾年中,諸多方法嘗試利用影像偵測來偵測每次治療可能出現的位移差異,但由於解析度僅512×512像素且演算程式複雜導致運算過慢,偵測效率不佳,無法滿足現階段臨床放射線治療之品質要求。近年來,醫療院所常見之VARIAN紅外線位移偵測裝置設備,可達每秒25 FPS與精密度2mm 的要求,但對於日漸普及之重粒子加速器要求之1mm 精密度仍無法滿足。 本文所提出之ScanNet演算法不僅能將解析度提升至1600×1200像素,又能達到每秒 25 FPS 的位移偵測目的,並可在治療過程中辨識出1mm的移位,符合重粒子加速器對於位移差異偵測之要求。且本文之建置籌本控制在8萬元以下,以降低病患自費項目之負擔。
Radiation therapy is a major modality of cancer treatment. To achieve good local control, high radiation dose must be precisely focused to the target region. A geometric missing during radiotherapy could result in tumor regrowth or severe damage of the neighboring normal tissues. To prevent geometric missing, a real-time monitoring the patient movement becomes very important. Using image or infrared technology to detect the displacement has been proposed. However, Promoting the current detection efficiency is too poor to fulfill clinical requirement due to the low resolution of pixels and slow complex calculation. To overcome these problems, this study proposed ScanNet algorithm to monitor the patient breathing, cough, or other movement. ScanNet algorithm could detect 1 mm shift under the resolution of 1600 × 1200 pixels in real-time and to meet the quality requirement of modern radiotherapy for monitoring displacement.