AI-based digital histopathology for perihilar cholangiocarcinoma: A step, not a jump.

Introduction: Perihilar cholangiocarcinoma (PHCC) is a rare malignancy with limited survival prediction accuracy. Artificial intelligence (AI) and digital pathology advancements have shown promise in predicting outcomes in cancer. We aimed to improve prognosis prediction for PHCC by combining AI-based histopathological slide analysis with clinical factors. Methods: We retrospectively analyzed 317 surgically treated PHCC patients (January 2009-December 2018) at the University Hospital of Essen. Clinical data, surgical details, pathology, and outcomes were collected. Convolutional neural networks (CNN) analyzed whole-slide images. Survival models incorporated clinical and histological features. Results: Among 142 eligible patients, independent survival predictors were tumor grade (G), tumor size (T), and intraoperative transfusion requirement. The CNN-based model combining clinical and histopathological features demonstrates proof of concept in prognosis prediction, limited by histopathological complexity and feature extraction challenges. However, the CNN-based model generated heatmaps assisting pathologists in identifying areas of interest. Conclusion: AI-based digital pathology showed potential in PHCC prognosis prediction, though refinement is necessary for clinical relevance. Future research should focus on enhancing AI models and exploring novel approaches to improve PHCC patient prognosis prediction.

Lymph Node Staging in Perihilar Cholangiocarcinoma: The Key to the Big Picture.

Klatskin tumors have a bad prognosis despite aggressive therapy. The role and extent of lymph node dissection during surgery is a matter of discussion. This retrospective study analyzes our current experience of surgical treatments in the last decade. Patients and Methods: A retrospective single-center analysis of patients (n = 317) who underwent surgical treatment for Klatskin tumors. Univariable and multivariable logistic regression and Cox proportional analysis were performed. The primary endpoint was to investigate the role of lymph node metastasis for patient survival after complete tumor resection. The secondary endpoint was the prediction of lymph node status and long-term survival from preoperatively available parameters. Results: In patients with negative resection margins, a negative lymph node status was the prognosis-determining factor with a 1-, 3-, and 5-year survival rate of 87.7%, 37%, and 26.4% compared with 69.5%, 13.9%, and 9.3% for lymph-node-positive patients, respectively. Multivariable logistic regression for complete resection and negative lymph node status demonstrated only Bismuth type 4 (p = 0.01) and tumor grading (p = 0.002) as independent predictors. In multivariate Cox regression analysis, independent predictors of survival after surgery were the preoperative bilirubin level (p = 0.03), intraoperative transfusion (p = 0.002), and tumor grading (G) (p = 0.001). Conclusion: Lymph node dissection is of utmost importance for adequate staging in patients undergoing surgery for perihilar cholangiocarcinoma. In spite of extensive surgery, long-term survival is clearly associated with the aggressiveness of the disease.