基于空间特征强化与贝叶斯不确定感知优化的卵巢癌预后预测模型

Ovarian Cancer Prognosis Prediction Model Based on Spatial Feature Enhancement and Bayesian Uncertainty Perception Optimization

摘要

卵巢癌是一种早期症状不明显且病灶形态复杂的妇科恶性肿瘤，卵巢肿瘤影像能够反映病灶的形态、结构及组织学信息，深入分析其特征可以提升卵巢癌预后预测的准确性。现有方法在处理卵巢肿瘤影像时，往往忽略局部空间、结构与纹理信息的协同作用，缺乏对病灶不确定信息的整体感知，因而模型的预测能力和稳定性仍有待提升。本文提出了一种基于空间特征强化与贝叶斯不确定感知优化的卵巢癌预后预测模型 SPT-BUAT（Spatial-Structure-Texture Pseudo-Temporal Bayesian Uncertainty-Aware Transformer），该模型采用空间-结构-纹理协同强化增强影像局部与全局特征表达，通过 BUAT 进行病灶不确定信息的估计与优化。实验结果表明，本文模型 SPT-BUAT 对卵巢癌预后预测的性能优于现有方法，可以对模糊边界与灰度异常区域进行更准确的判别。进一步对 SPT-BUAT 进行了可视化分析，验证了模型的关注区域与实际病灶位置的一致性，为深度学习在卵巢肿瘤影像分析中的应用研究提供了新的思路。

Abstract

Ovarian cancer is a kind of malignant gynecological tumor with inconspicuous early symptoms and complex lesion morphology. Ovarian tumor images can reflect the morphological, structural and histological information of lesions, and in-depth analysis of these features can improve the accuracy of ovarian cancer prognosis prediction. Existing methods tend to ignore the synergistic effect of local spatial, structural and texture information when processing ovarian tumor images, and lack holistic perception of uncertain lesion information, leaving room for improvement in the predictive performance and stability of models. This paper proposes an ovarian cancer prognosis prediction model named SPT-BUAT (Space-Structure-Texture Pseudo-Temporal Bayesian Uncertainty-Aware Transformer) based on spatial feature enhancement and Bayesian uncertainty-aware optimization. The model leverages space-structure-texture complementary model (SST-CM) to boost the representation of local and global image features, and employs BUAT to estimate and optimize the uncertain information of lesions. Experimental results demonstrate that the proposed SPT-BUAT model outperforms existing methods in ovarian cancer prognosis prediction and achieves more accurate discrimination of blurred boundaries and gray abnormal regions. Further visual analysis of SPT-BUAT verifies the consistency between the attention regions of the model and the actual lesion locations, which provides a new research perspective for the application of deep learning in ovarian tumor image analysis.

关键词

空间-结构-纹理协同强化模块 / 伪时序建模 / 贝叶斯不确定感知 / 卵巢癌预后预测 / 卵巢肿瘤影像

Key words

space-structure-texture complementary model / pseudo time series modeling / Bayesian uncertainty perception / prognostic prediction of ovarian cancer / ovarian tumor imaging

参考文献

[1] CHEN X, HUO X F, WU Z, et al. Advances of artificial intelligence application in medical imaging of ovarian cancers[J].Chinese Medical Sciences Journal, 2021, 36(3):196-203.

[2] JIANG R, FENG Y, CHEN Y, et al. Surgery versus no surgery in platinum-sensitive relapsed ovarian cancer: Final overall survival analysis of the SOC-1 randomized phase 3 trial[J].Nature Medicine, 2024, 30(8):2181-2188.

[3] ZENG S, WANG L X, YANG H. Radiomics and radiogenomics: Extracting more information from medical images for the diagnosis and prognostic prediction of ovarian cancer[J].Military Medical Research, 2024, 11(1):77.

[4] DENG Q Y, LU L P, LIU J L, et al. Diagnostic value of tumor markers combined with different imaging protocols for ovarian tumors[J].European Journal of Gynaecological Oncology, 2025, 46(4):89-95.

[5] KUNISHIMA A, INABA D, IYOSHI S, et al. Application of multimodal integration to develop preoperative diagnostic models for borderline and malignant ovarian tumors[J].Scientific Reports, 2025, 15(1):37114.

[6] XIE Q Y, HUANG J N, SUN J Y, et al. SMoFFI-SegFormer: A novel approach for ovarian tumor segmentation based on an improved SegFormer architecture[J].Frontiers in Oncology, 2025, 15:1555585.

[7] GHANTASALA G S P, DILIP K, VIDYULLATHA P, et al. Enhanced ovarian cancer survival prediction using temporal analysis and graph neural networks[J].BMC Medical Informatics and Decision Making, 2024, 24(1):299.

[8] HU J Q, CUI Z M, ZHANG X, et al. Uncertainty-aware refinement framework for ovarian tumor segmentation in CECT volume[J].Medical Physics, 2023, 51(4):2678-2694.

[9] GONZÁLEZ J A, WROBEL J, VANDEKAR S, et al. Analyzing spatial point patterns in digital pathology: Immune cells in high-grade serous ovarian carcinomas[J].The American Statistician, 2025, 79(3):355-370.

[10] DELAGDO-BARRIGA K, MEDINA C, GOMEZ-QUILES L, et al. CT Enterography for preoperative evaluation of peritoneal carcinomatosis index in advanced ovarian cancer[J].Journal of Clinical Medicine, 2022, 11(3):476.

[11] RUNDO L, BEER L, URSPRUNG S, et al. Tissue-specific and interpretable sub-segmentation of whole tumour burden on CT images by unsupervised fuzzy clustering[J].Computers in Biology and Medicine, 2020, 120:103751.

[12] SIAHPOOSH M, IMANI M, GHASSEMIAN H. Ovarian tumor ultrasound image segmentation with deep neural networks[C]//2024 13th Iranian/3rd International Machine Vision and Image Processing Conference (MVIP). Piscataway, NJ: IEEE, 2024:1-5.

[13] CHO N S, LE V L, SANVITO F, et al. Digital "flipbooks" for enhanced visual assessment of simple and complex brain tumors[J].Neuro-Oncology, 2024, 26(10):1823-1836.

[14] ALSALLAL M, AHMED H H, KAREEM R A, et al. Enhanced lung cancer subtype classification using attention-integrated DeepCNN and radiomic features from CT images: A focus on feature reproducibility[J].Discover Oncology, 2025, 16(1):336.

[15] XIONG X, HONG R, XU F, et al. Quantitative assessment of bone marrow infiltration and characterization of tumor burden using dual-layer spectral CT in patients with multiple myeloma[J].Radiology and Oncology, 2024, 58(1):43-50.

[16] GUI H T, SU T, PANG Z Y, et al. Diagnosis of breast cancer with strongly supervised deep learning neural network[J].Electronics, 2022, 11(19):3003.

[17] AHMAD N, CHEN Y T. Eff-SAM: SAM-based efficient method for brain tumor segmentation in multimodal 3D MRI scans[J].IEEE Access, 2025, 13:90063-90087.

[18] SEONI S, JAHMUNAH V, SALVI M, et al. Application of uncertainty quantification to artificial intelligence in healthcare: A review of last decade (2013-2023)[J].Computers in Biology and Medicine, 2023, 165:107441.

[19] 高佳希, 黄海燕. 基于 TF-IDF 和多头注意力 Transformer 模型的文本情感分析 [J].华东理工大学学报 (自然科学版), 2024, 50(1):129-136.

[20] 周腊珍, 陈红池, 李秋霞, 等. 基于 Transformer 深度学习模型在医学图像分割中的研究进展 [J].中国生物医学工程学报, 2024, 43(4):467-476.

[21] CLAESSENS C H B, SCHULTZ E W R, KOCH A, et al. Multi-center ovarian tumor classification using hierarchical transformer-based multiple-instance learning[J].Cancer Prevention, Detection, and Intervention, 2025, 15199:3-13.

[22] AELGANI V, VADLAKONDA D. A novel interpretable regularized cnn with a modified xlnet transformer for segmenting and classifying the ovarian cancer[J].Multimedia Tools and Applications, 2024, 84(12):1-28.

[23] ZHANG Y, TAN J, WANG J P, et al. Are CT and MRI useful tools to distinguish between micropapillary type and typical type of ovarian serous borderline tumors?[J].Abdominal Radiology, 2021, 46(7):1-11.

[24] KIM H J, LEE S Y, SHIN Y R, et al. The value of diffusion-weighted imaging in the differential diagnosis of ovarian lesions: A meta-analysis[J].PLOS One, 2017, 11(2):e0149465.

[25] YIN J, HUANG Y L, LU C H, et al. Uncertainty-based saltwater intrusion prediction using integrated Bayesian machine learning modeling (IBMLM) in a deep aquifer[J].Journal of Environmental Management, 2024, 354:120252.

[26] WANG X H, XU C, GRZEGORZEK M, et al. Habitat radiomics analysis of PET/CT imaging in high-grade serous ovarian cancer: Application to Ki-67 status and progression-free survival[J].Frontiers in Physiology, 2022, 13:948767.

[27] MANFREDI M, MORABITO S, FOURNIER Q, et al. Computed tomographic findings in dogs with ovarian tumors: A tortuous ovarian artery consistently identifies ovarian origin in complex abdominal masses[J].Veterinary Radiology & Ultrasound, 2025, 66(1):e13476.

[28] TANABE K, IKEDA M, HAYASHI M, et al. Comprehensive serum glycopeptide spectra analysis combined with artificial intelligence (CSGSA-AI) to diagnose early-stage ovarian cancer[J].Cancers, 2020, 12(9):2373.

[29] KODIPALLI A, FERNANDES S L, GURURAJ V, et al. Performance analysis of segmentation and classification of CT-Scanned ovarian tumours using U-Net and deep convolutional neural networks[J].Diagnostics, 2023, 13(13):2282.

[30] GUHA S, KODIPALLI A, FERNANDES S L, et al. Explainable AI for interpretation of ovarian tumor classification using enhanced ResNet50[J].Diagnostics, 2024, 14(14):1567.

[31] ZHANG H Y, HU D Q, DUAN H L, et al. A novel deep learning approach to extract Chinese clinical entities for lung cancer screening and staging[J].BMC Medical Informatics and Decision Making, 2021, 21(Suppl 2):214.

[32] ZHENG Y W, WANG F, ZHANG W X, et al. Preoperative CT-based deep learning model for predicting overall survival in patients with high-grade serous ovarian cancer[J].Frontiers in Oncology, 2022, 12:986089.

[33] QI R, WU W N, LU Z T, et al. Decision-level fusion classification of ovarian CT benign and malignant tumors based on radiomics and deep learning of dual views[J].IEEE Access, 2024, 12:102381-102395.

[34] MASMALI I, AHMAD A, AZEEM M, et al. TOPSIS method based on intuitionistic fuzzy soft set and its application to diagnosis of ovarian cancer[J].International Journal of Computational Intelligence Systems, 2024, 17(1):161.

[35] XIE P, DING J, WANG X G. Leveraging external aggregated information for the marginal accelerated failure time model[J].Statistics in Medicine, 2024, 43(27):5203-5216.

[36] ALSHDAIFAT E H, GHARAIBEH H, SINDIANI A M, et al. Hybrid vision transformer and Xception model for reliable CT-based ovarian neoplasms diagnosis[J].Intelligence-Based Medicine, 2025, 11:100227.

[37] OPENAI H. Hello GPT-5[EB/OL]. (2025-12-17). https://openai.com/zh-Hans-CN/index/introducing-gpt-5/.

[38] LEE C , ZAME W , YOON J , et al. DeepHit: A deep learning approach to survival analysis with competing risks[J].Proceedings of the AAAI Conference on Artificial Intelligence, 2018, 32(1):2314-2321.

[39] XIN Q, CHEN X X, WANG C. Novel convex model-based approach for data-driven fault diagnosis considering uncertainty[J].Reliability Engineering and System Safety, 2026, 266(PA):111714.

基金资助

山西省自然科学基金(202203021211121)