Comparative evaluation study on automated contouring methods using deep learning technique

Accurate delineation of target volumes and organs-at-risk (OARs) is critical in radiation therapy planning. Manual contouring, while standard, is time-consuming and subject to variability. This study compared manual contouring with an AI-based deep learning auto-contouring software in 30 patients from head & neck, abdomen, and pelvis regions. Key OARs including parotid glands, kidneys, spleen, bladder, rectum, and femoral heads were analyzed using conformity index (CI), mean distance to conformity, Dice similarity coefficient, inclusiveness index, sensitivity index, and error volume histogram. Automated contouring reduced contouring time (5–10 min vs. approximately 1 h) while maintaining comparable geometric accuracy. The kidneys showed the highest CI (0.93), while parotid glands demonstrated lower values (0.76–0.77). These results suggest AI-based auto-contouring is a clinically viable approach, particularly in abdominal and pelvic regions, though partial manual adjustments may still be required in complex anatomical sites such as the head and neck. In addition to the quantitative improvements, the findings highlight the potential of AI-based contouring to enhance workflow standardization and improve consistency in clinical practice.