Predicting early in treatment whether a tumor is likely to be responsive is a difficult yet important task for personalized patient care. Studies have shown that multimodal biomarkers could provide complementary information and lead to more accurate treatment outcome prognosis than unimodal biomarkers. However, in most cases, the usage of multimodal biomakers faces up to the challenges of the heterogeneity and incompleteness of multimodal data and the small-sized training dataset. An efficient strategy aiming to address these challenges remains largely lacking. In this work, a modulized framework was proposed to employ multimodal biomarkers for cancer treatment outcome prediction, and enable the seamless integration, validation and comparison of various algorithms. The proposed framework contained four modules of synthetic data generation, deep feature extraction, multimodal feature fusion, and classification.
The proposed framework is shown in Figure 1., which includes four modules of synthetic data generator, deep feature extraction, multimodal feature fusion, and classification.
The proposed method was tested on a set of 305 oropharyngeal squamous cell carcinoma (OPSCC) patients for patient stratification with low- and high-risk of treatment failures. Receiver operating characteristics (ROC) was used to evaluate the performnace of the proposed method. The method was compared with 4 other competing methods. Their ROC performances are shown in the Figure 2.
1. Maliazurina Saad, Shenghua He, Wade Thorstad, Hiram Gay, Su Ruan, Xiaowei Wang, and Hua Li*,“Learning-based Cancer Treatment Outcome Prognosis using Multimodal Biomarkers”. (IEEE Transactions on Radiation and Plasma Medical Sciences 2021 under review)
2. Maliazurina Saad, Shenghua He, Wade Thorstad, Hiram Gay, Xue Wu, Su Ruan, Xiaowei Wang, and Hua Li*, “Leveraging Incomplete Multimodal Biomarkers for Cancer Treatment Outcome Prediction”, AAPM 2020.
3. Maliazurina Saad, Shenghua He, Wade Thorstad, Hiram Gay, Xue Wu, Su Ruan, Xiaowei Wang, and Hua Li*, “Multimodal Biomarkers for Cancer Treatment Outcome Prediction by Use of Deep Learning and Canonical Correlation Analysis”, AAPM 2020.