Room: Exhibit Hall
Purpose: Manual contouring of gross tumor volumes (GTV) is a crucial and time-consuming process in lung cancer radiotherapy. This study aims to develop a simple deep learning based auto segmentation algorithm to segment lung tumors on T2 weighted CT images.
Methods: CT scans of 542 non-small cell lung cancer (NSCLC) patients were enrolled in this study, including 247 early stage lung cancer (stage I) and 295 early stage lung cancer (stage II and stage III) treated with radiotherapy. 2D U-net similar network was established as a training model. Data were randomly separated into training (90%) and validation (10%) datasets. Four indices were calculated to evaluate the similarity of automated and manual segmentation, including Hausdorff distance (HD), average surface distance (ASD), Dice index (DSC), and Jaccard index (JSC).
Results: In all cases, each prediction of automatically delineation of lung cancer took approximately 1 min on a PC workstation with TITAN X Pascal and 98 GB of memory. The method was trained based on 542 non-small cell lung cancer (NSCLC) patientsâ€™ data and was twas tested on 44 lung nodule CT validation datasets. The The validation tests resulted in Dice index (DSC), Jaccard index (JSC), Hausdorff distance (HD), average surface distance (ASD) (meanÂ±SD) were 0.67Â±0.23, 0.54Â±0.22, 93.22Â±83.10mm and 12.44Â±18.60mm for validation dataset.
Conclusion: This study showed that a simple deep learning neural network can perform segmentation for lung cancer based on CT.