Training, validation, and clinical implementation of a deep-learning segmentation model for radiotherapy of loco-regional breast cancer

Abstract

Aim: To train and validate a comprehensive deep-learning (DL) segmentation model for loco-regional breast cancer with the aim of clinical implementation.

Methods: DL segmentation models for 7 clinical target volumes (CTVs) and 11 organs at risk (OARs) were trained on 170 left-sided breast cancer cases from two radiotherapy centres in Norway. Another 30 patient cases were used for validation, which included the evaluation of Dice similarity coefficient and Hausdorff distance, qualitative scoring according to clinical usability, and relevant dosimetric parameters. The manual inter-observer variation (IOV) was also evaluated and served as a benchmark. Delineation of the target volumes followed the ESTRO guidelines.

Results: Based on the geometric similarity metrics, the model performed significantly better than IOV for most structures. Qualitatively, no or only minor corrections were required for 14% and 71% of the CTVs and 72% and 26% of the OARs, respectively. Major corrections were required for 15% of the CTVs and 2% of the OARs. The most frequent corrections occurred in the cranial and caudal parts of the structures. The dose coverage, based on D98 > 95%, was fulfilled for 100% and 89% of the breast and lymph node CTVs, respectively. No differences in OAR dose parameters were considered clinically relevant. The model was implemented in a commercial treatment planning system, which generates the structures in 1.5 min.

Conclusion: Convincing results from the validation led to the decision of clinical implementation. The clinical use will be monitored regarding applicability, standardization and efficiency.

Keywords: Artificial intelligence; Auto-segmentation; Breast cancer; Deep learning; Radiation therapy.