Poster S1 - Investigating the Correlation Between Treatment Response and Overall Survival Rates in Pancreatic Ductal Adenocarcinoma Patients

Introduction: Pancreatic ductal adenocarcinoma (PDAC), known for its high aggressiveness, ranks as the fourth most prevalent cause of death in United States. The 5-year relative survival rate for PDAC stands at a mere 13% and its poor prognosis is linked to the lack of effective treatment options. Patients with metastatic PDAC who received FOLFIRINOX AND gemcitabine combined with nab-paclitaxel showed only a slight improvement in survival, increasing life expectancy by approximately 10 to 11 months. The utilization of quantitative analyses of histological features such as prognostic biomarker expression can provide valuable insight on the response to treatment therapies in relation to imaging results. Previous researchers investigated cure rate models to gain insight on the survival rates for osteosarcoma patients and discovered the histological response strongly predicts cure likelihood in high-grade osteosarcoma patients with 55% cure rates for good responders and 29% for poor responders. The aim of this current study was to examine how treatment response impacts overall survival in PDAC patients.

Materials and

Methods: The treatment response to cell ratio in the presence of four different biomarkers was analyzed using MATLAB for the first cohort (n=12). An ROC curve was produced to display cut offs. Based on the cut offs, a transfer learning immunohistochemical (IHC) imaging model was developed to classify images by overall survival in cohort 2 (n=58).

Results, Conclusions, and Discussions:

Results: The ROC curve for CD45+ had a sensitivity of 63.41% and a specificity 82.14% and IL10 had a sensitivity of 60.98% and a specificity 89.29%. The confusion matrix from the transfer learning IHC imaging model validation results with 53.2% of the predictions being correct. The low accuracy could be due to lack of data and training time which will be adjusted in future work. The number and percentages of the diagonal cells correspond to the observations correctly classified, while the off-diagonal cells correspond to incorrectly classified observations. The column on the far right shows the percentages of the predicted class that are correctly and incorrectly classified. The row at the bottom shows the percentages of each class that are correctly and incorrectly classified.

Conclusion: While the ROC curve analysis shows that CD45+ and IL10 markers have potential in diagnostic applications with their moderate sensitivity and specificity, the current transfer learning IHC imaging model requires significant enhancements. The model's accuracy of 53.2% is insufficient for reliable diagnostic use. To improve this, future work should focus on increasing the dataset size and extending the training duration. These adjustments are expected to enhance the model's ability to accurately classify observations, thereby improving its overall performance and diagnostic utility.

Discussion: The ROC curve analysis indicates moderate diagnostic performance for CD45+ and IL10, with CD45+ showing reasonable sensitivity and specificity, and IL10 having slightly higher specificity. However, the IHC imaging model's validation results reveal a low accuracy of 53.2%, likely due to limited data and training time. The confusion matrix shows

Acknowledgements (Optional):