Multimodal prediction for radiotherapy-induced hematologic toxicity in rectal cancer patients.
Authors
Affiliations (12)
Affiliations (12)
- School of Biomedical Engineering, Beijing Key Laboratory of Clinical Engineering Solutions for Mental Health, Capital Medical University, Beijing, China. [email protected].
- School of Computer Science and Technology, Xidian University, Xi'an, China.
- Department of Radiation Oncology & Therapy, The First Hospital of Jilin University, Changchun, China.
- Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Department of Radiation Oncology, Peking University Cancer Hospital & Institute, Beijing, China.
- Department of Radiation Oncology, Peking University Cancer Hospital (Inner Mongolia Campus) & Affiliated Cancer Hospital of Inner Mongolia Medical University, Huhhot, China.
- Department of Radiology, Beijing Children's Hospital, Capital Medical University, National Center for Children's Health, Beijing, China.
- Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Department of Radiation Oncology, Peking University Cancer Hospital & Institute, Beijing, China. [email protected].
- Department of Radiation Oncology, Peking University Cancer Hospital (Inner Mongolia Campus) & Affiliated Cancer Hospital of Inner Mongolia Medical University, Huhhot, China. [email protected].
- School of Biomedical Engineering, Beijing Key Laboratory of Clinical Engineering Solutions for Mental Health, Capital Medical University, Beijing, China.
- Department of Radiology and Radiological Sciences, Johns Hopkins University, Baltimore, MD, USA.
- Department of Diagnostic Imaging, Brown University, Providence, RI, USA.
- School of Computer Science and Technology, Xidian University, Xi'an, China. [email protected].
Abstract
Early identification of acute hematologic toxicity (HT) in locally advanced rectal cancer (LARC) patients undergoing radiotherapy is crucial for optimizing clinical outcomes. Here, we retrospectively collected multi-center LARC patients (n = 464, n = 56, and n = 79) with complete CT images, dose maps, hematologic biomarkers, and demographic information. A Transformer-based multimodal fusion model was constructed to combine the visual and non-visual representation features for HT prediction, and the study also testified to the modality-specific and region-specific contributions to HT. The multimodal fusion model achieved a state-of-the-art HT prediction performance in LARC patients: with an area under the curve (AUC) of 0.828 (95% confidence interval [CI]: 0.820-0.835), 0.757 (95% CI: 0.750-0.766), and 0.756 (95% CI: 0.752-0.762) in internal and two external testing datasets. The initial hematologic biomarkers were the best unimodal risk indicator, while the planning target volume served as the most sensitive region. The study confirms the sole and combined contributions of each modality to the radiotherapy-induced HT in LARC patients, and the multimodal fusion model shows promising interpretability and generalization for HT occurrence, which offers valuable insights to optimize personalized treatment plans for high-risk patients.