Patient-specific three-dimensional dose distribution prediction via deep learning for prostate cancer therapy: Improvement with the structure loss. (March 2023)

• Record Type: Journal Article
• Authors: Koike, Yuhei; Takegawa, Hideki; Anetai, Yusuke; Ohira, Shingo; Nakamura, Satoaki; Tanigawa, Noboru
• Is Part Of: Physica medica. Volume 107(2023)
• Publication Date: 2023-03
• Subjects: Deep learning -- Patient-specific dose distribution -- VMAT -- Prostate cancer; Medical physics -- Periodicals; Biophysics -- Periodicals; Biophysics -- Periodicals; Imagerie médicale -- Périodiques; Radiothérapie -- Périodiques; Rayons X -- Sécurité -- Mesures -- Périodiques; Physique -- Périodiques; Médecine -- Périodiques; 610.153
• DOI: 10.1016/j.ejmp.2023.102544
• Languages: English
• ISSNs: 1120-1797

Abstract

Highlights: A deep learning-based method was developed for VMAT dose distribution. A structure loss (Lstruct ) was proposed to calculate errors per input contour labels. Model performance of Lstruct was compared with a standard L2 loss. Lstruct model yielded more accurate dose distributions and outperformed L2 loss. Abstract: Purpose: Deep learning (DL)-based dose distribution prediction can potentially reduce the cost of inverse planning process. We developed and introduced a structure-focused loss (Lstruct ) for 3D dose prediction to improve prediction accuracy. This study investigated the influence of Lstruct on DL-based dose prediction for patients with prostate cancer. The proposed Lstruct, which is similar in concept to dose–volume histogram (DVH)-based optimization in clinical practice, has the potential to provide more interpretable and accurate DL-based optimization. Methods: This study involved 104 patients who underwent prostate radiotherapy. We used 3D U-Net-based architecture to predict dose distributions from computed tomography and contours of the planning target volume and organs-at-risk. We trained two models using different loss functions: L2 loss and Lstruct . Predicted doses were compared in terms of dose–volume parameters and the Dice similarity coefficient of isodose volume. Results: DVH analysis showed that the Lstruct model had smaller errors from the ground truth than the L2 model. The Lstruct model achieved more consistent dose distributions than the L2 model, with errors close to zero. The isodose Dice score of the Lstruct model was greater than that of the L2 model by >20% of the prescribed dose. Conclusions: We developed Lstruct using labels of inputted contours for DL-based dose prediction for prostate radiotherapy. Lstruct can be generalized to any DL architecture, thereby enhancing the dose prediction accuracy.