📄 中文摘要
利用AI模型预测前列腺癌根治术（RP）后的生化复发（BCR），该模型以诊断性前列腺活检玻片作为输入数据。BCR是侵袭性前列腺癌预后不良的替代标志物，但现有预测工具的精确度有限。通过在STHLM3队列的676例患者诊断性前列腺活检玻片数据上训练AI模型，该模型结合了基础模型和基于注意力机制的多实例学习方法，旨在预测患者特异性的BCR风险。为了评估模型的泛化能力，研究人员在三个独立的根治性前列腺切除术（RP）样本数据集上进行了外部验证。这些数据集来自不同的机构和患者群体，确保了模型在不同临床背景下的适用性。模型训练过程中，主要关注从活检图像中提取与肿瘤形态、组织结构和细胞特征相关的潜在生物标志物。
📄 English Summary
AI-based Prediction of Biochemical Recurrence from Biopsy and Prostatectomy Samples
An AI-based model predicts biochemical recurrence (BCR) following radical prostatectomy (RP), utilizing diagnostic prostate biopsy slides as input. BCR serves as a surrogate marker for aggressive prostate cancer with adverse outcomes, yet current prognostic tools exhibit limited precision. The AI model was trained on diagnostic prostate biopsy slides from the STHLM3 cohort, comprising 676 patients. It incorporates foundation models and attention-based multiple instance learning to predict patient-specific BCR risk. To assess the model's generalizability, external validation was performed across three independent radical prostatectomy (RP) sample datasets. These datasets originated from diverse institutions and patient populations, ensuring the model's applicability across various clinical contexts. During model training, the primary focus was on extracting latent biomarkers related to tumor morphology, tissue architecture, and cellular characteristics from biopsy images. Through a deep learning architecture, the model identifies microscopic patterns that are challenging to quantify using conventional pathology, and these patterns may be closely associated with disease recurrence. The multiple instance learning approach enables the model to learn key features from multiple biopsy tissue sections without precise annotations, enhancing its robustness. The attention mechanism allows the model to focus on the most predictively valuable regions within the images, thereby improving prediction accuracy and interpretability. Ultimately, this AI model aims to provide clinicians with a more precise tool for identifying high-risk patients, thereby guiding postoperative surveillance and adjuvant treatment decisions, with the goal of improving treatment outcomes and quality of life for prostate cancer patients.