Quality Assessment of Commercial Contouring Software based on Convolutional Neural Networks: Implications for Clinical Practice
Year: 2025
Type: article
Abstract: The treatment planning for radiotherapy, particularly the delineation of pelvic structures, is a complex and time-consuming task, demanding considerable expertise from physicians. The development of automation tools for these procedures could potentially reduce the workload and improve treatment planning efficiency. The goal of this study is to evaluate the efficacy of the AutoContour software, a commercial auto-segmentation tool utilizing convolutional neural networks, in delineating organs at risk during cervical cancer radiotherapy. We evaluated the performance of the AutoContour software on fifteen pelvic CT scans from cervical cancer patients previously treated at our institution. This study compared the automatically generated structures with those manually delineated by two experienced radiation oncologists. The comparison employed the Dice Similarity Coefficient (DSC) to assess the quality of structures, and we also measured the time savings achieved by using the auto-segmentation tool. The study found that most of the structures delineated by the AutoContour software closely matched those contoured by the radiation oncologists, with DSC values greater than 0.80, indicating high similarity. However, the bowel bag showed lower similarity, which could be attributed to interobserver variability among the physicians themselves. The use of AutoContour resulted in a reduction of up to 25.86 minutes in the time required per patient for structure delineation, demonstrating substantial efficiency gains without compromising the quality of the contours. The AutoContour software streamlines the delineation process in cervical cancer radiotherapy planning, maintaining high-quality output with minimal need for adjustments. These results suggest that the integration of this auto-segmentation tool could considerably decrease the specialized workload, enhancing the overall efficiency of clinical workflows in radiation oncology departments. This automation not only saves time but also reduces the potential for human error, promising more consistent and reliable treatment planning.
Institutions: Hospital Universitário de Brasília, Universidade de Brasília
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4408166120