Abstract: Ovarian cancer is a gynecological malignant tumor with unclear early symptoms and complex lesion morphology. It has the characteristics of insidious onset, rapid progression, and poor prognosis, and is one of the main diseases that threaten women's life and health. Ovarian tumor imaging can reflect the morphology, structure, and histological information of the lesion. In depth analysis of its characteristics not only helps to evaluate the staging and grading of the tumor, but also provides reference for individualized treatment plans in clinical practice. With the development of medical image analysis and artificial intelligence technology, automated image analysis methods based on deep learning have become the main research object in the diagnosis and prognosis prediction of ovarian cancer. Existing methods often overlook the synergistic effect of local spatial, structural, and texture information when processing ovarian tumor images, lacking overall perception of uncertain lesion information, resulting in the need to improve the predictive ability and stability of the model. Therefore, this article proposes an ovarian cancer prognosis prediction model SPT-BUAT based on spatial feature enhancement and Bayesian uncertainty perception optimization. The model uses spatial structure texture collaborative enhancement to enhance local and global feature expression in images, and estimates and optimizes lesion uncertainty information through Bayesian uncertainty perception optimization Transformer. The experimental results show that the SPT-BUAT model proposed in this paper performs better than existing methods in predicting the prognosis of ovarian cancer, and can more accurately distinguish between fuzzy boundaries and gray abnormal areas. This article further visualized SPT-BUAT and verified the consistency between the model's attention area and the actual lesion location, providing new ideas for the application of deep learning in ovarian tumor imaging analysis.