Radiomics-guided Automatic Delineation for Clinical Target Volume of Endometrial Cancer: Limited-sample Multi-center Study.

Authors

Abstract

The inconsistent delineation style of clinical target volume (CTV) in postoperative pelvic radiotherapy of endometrial carcinoma across different centers is challenging for deep learning-based segmentation due to different definitions of internal target areas. This study aims to develop an effective method to address multi-institutional variations in CTV delineation, even under the constraints of scarce data availability. A total of 207 simulated CT cases from endometrial cancer patients across five centers were retrospectively collected. Within each center, the data were divided into support, query, and test sets. Each center was sequentially designated as the target center for fine-tuning and testing, while the remaining four centers were used for model training to validate the superiority of the proposed method. In addition, 26 cases from an external center were used exclusively for fine-tuning and testing. Radiomics features were extracted to analyze the differences in CTV delineation and image among the centers. A Random Forest Classifier (RFC) was trained to identify the most important radiomics features. Using these features as guidance, a Model-Agnostic Meta-Learning (MAML) strategy was applied to pre-train a 3D U-Net model (MAML-r), which was subsequently fine-tuned on each target center's data. The performance of the proposed MAML-r method was compared against direct 3D U-Net training and transfer learning approaches. Evaluation metrics included the Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD95), and Average Symmetric Surface Distance (ASSD), supplemented by qualitative assessment from clinical experts using a four-point scoring system. The 8 most important features were identified from a total of 107 radiomics features, which showed significant differences across centers (p < 0.01). The MAML-r model yielded meaningful results, achieving a DSC a mean DSC of 0.818 ± 0.058, a mean HD95 of 9.314 ± 3.648 mm, and a mean ASSD of 2.772 ± 1.090 mm. It also earned an average blinded expert evaluation score of 3.24, significantly outperforming all other models. Notably, improved performance was observed on the external test cohort, with corresponding values of 0.886 ± 0.012, 5.203 ± 1.435 mm, and 1.512 ± 0.334 mm, respectively. Furthermore, the MAML-r model achieved the shortest CTV modification time of 3.8±1.2 minutes. Given the variations in CTV contouring styles across different centers with limited samples, the MAML-r model demonstrates superior performance and adaptability compared to other models. This study introduces a novel radiomics-guided MAML framework for few-shot, multi-centric CTV segmentation tasks in postoperative pelvic radiotherapy for endometrial carcinoma, significantly mitigating performance degradation caused by inter-institutional delineation style variations and data scarcity. The proposed approach thus offers a promising solution to these persistent clinical challenges.

Tags

Topics