OCProNet: Ovarian Cancer Prognosis Prediction Network Based on Hybrid CNN and ViT
Kai Zou, Qianlan Yang, Dawei Dai · 2024
With the rapid development of medical imaging technology and deep learning, prognostic outcome prediction has become an important area of medical research. Prognostic evaluation of ovarian cancer is crucial for patient rehabilitation management. In previous studies, various advanced deep convolutional neural network (CNN) models have been used for prognostic assessment of different cancers. However, CNN models typically focus on learning local feature details for decision-making, often overlooking global features. In this study, we propose a hybrid algorithm based on Convolutional Neural Networks (CNNs) and Vision Transformers (ViT) to accurately predict three key indicators: Lymph Node Metastasis (LNM), Organ Involvement (Organ), and Relapse. First, CNNs are used to preliminarily extract features from medical images, while ViT captures long-range dependencies and global information. The combination of the two aims to improve the accuracy of cancer prognostic predictions. Experimental results demonstrate that the proposed method excels in predicting LNM, Organ Involvement, and Relapse, achieving an average accuracy of 82.54%, precision of 0.82, and recall of 0.82, significantly outperforming traditional models. This algorithm not only enhances the accuracy of prognostic predictions but also provides strong data support for personalized treatment planning.