Multi-branch Residual Network Applied to Predict the Three-Year Survival of Patients with Glioblastoma
- PDF / 1,386,128 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 80 Downloads / 175 Views
ORIGINAL ARTICLE
Multi‑branch Residual Network Applied to Predict the Three‑Year Survival of Patients with Glioblastoma Xue Fu1 · Chunxiao Chen1 · Dongsheng Li1 Received: 27 December 2019 / Accepted: 6 August 2020 © Taiwanese Society of Biomedical Engineering 2020
Abstract Purpose Glioblastoma is the most common primary brain tumor worldwide. Computer-aided survival prediction can provide a scientific basis for doctors to develop treatment plans to effectively avoid excessive treatment and waste of medical resources. Therefore, we used an end-to-end deep network based on radiomics to make survival predictions of patients with glioblastoma. Methods 360 magnetic resonance imaging images of glioblastoma patients were randomly selected from the TCIA database including T1 weighted and T2 weighted images. Based on the traditional residual network (ResNet), a two-branch residual network survival prediction (BSP) model was proposed to extract and learn the features from T1 and T2 images separately. Furthermore, considering the association of tumor area and whole brain tissues, a multi-branch residual network survival prediction (M-BSP) model was proposed, which can make full use of the features of the tumor image and the whole brain image. Results The classification accuracy of M-BSP model using different amounts of residual blocks on test sequence was 89%, 83% and 83%, respectively, demonstrating that the M-BSP model with two residual blocks performed better in prediction. Further, the survival analysis of the prediction results indicated that the M-BSP model can effectively classify patients into a high-risk group and a low-risk group. Conclusion The classification and prediction results demonstrated that the M-BSP model can attain superior classification results, which can assist doctors in making diagnostic decisions and developing treatment plans. Keywords Glioblastoma · Radiomics · Deep learning · Survival prediction
1 Introduction Glioblastoma is the most common primary brain tumor caused by carcinogenesis of brain and spinal glial cells. The survival prediction of glioblastoma patients can assist doctors in making medical plans to better allocate medical resources. Usually, doctors develop treatment plans based on patients’ clinical information, histopathological information and medical images. At present, some histopathological examinations may take a long time and cause injury to the human body. With the maturity of medical imaging technology and the development of radiomics, medical images are becoming more important in cancer diagnosis and treatment, * Chunxiao Chen [email protected] 1
Department of Biomedical Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China
and many researchers have shifted their attention to the study of survival prediction based on radiomics [1, 2]. It is hard to make accurate survival predictions of the cancer patients due to the difficulty of following up and incomplete image data. Machine learning methods are now widely used for survival
Data Loading...
