Abstract LB118: Machine learning-based tumor grading in pancreatic ductal adenocarcinoma: Exploring texture features for automated classification and clinical decision support

Abstract Pancreatic ductal adenocarcinoma (PDAC) is an aggressive malignancy and a leading cause of cancer-related death in the U.S. Due to late-onset symptoms, it often remains undiagnosed until advanced stages, resulting in poor prognosis. This study presents a machine learning approach to classify tumor grades based on texture features extracted from histological images, offering insights into prognosis and treatment decisions. A 2019 study by Qiu et al. demonstrated the effectiveness of machine learning-based CT texture analysis in predicting PDAC histopathological grades, achieving 86% accuracy, 78% sensitivity, and 95% specificity. Similarly, this work utilizes images from four tumor grades—Normal, Grade I, Grade II, and Grade III—obtained from Hematoxylin and Eosin and May-Grunwald-Giemsa staining. Texture features, including Gray-Level Co-occurrence Matrix (GLCM) properties, Local Binary Pattern (LBP) features, and Histogram of Oriented Gradients (HOG), were extracted to create a feature vector for each image. These vectors were used to train a Support Vector Machine (SVM) model with Error-Correcting Output Codes (ECOC) for multiclass classification, with hyperparameter optimization to improve model performance. Cross-validation was used to evaluate the model, yielding an accuracy of 53%. Although this accuracy is suboptimal, it represents a foundational step in automated PDAC tumor classification. The model could still serve as a useful tool in clinical practice, especially when used alongside other diagnostic methods or as a baseline for further improvements. Principal Component Analysis (PCA) was applied to visualize the feature distribution, and a confusion matrix was generated to assess classification performance. Results indicate that, despite the modest accuracy, the extracted texture features have potential for distinguishing between tumor grades, providing a starting point for automated classification and supporting clinical decision-making. The study introduces an innovative approach to PDAC tumor grading, addressing the urgent need for improved diagnostic tools. While further work is needed to optimize performance, this research sets the stage for future advancements that could impact clinical decision-making and patient outcomes. Citation Format: Miracle Thomas. Machine learning-based tumor grading in pancreatic ductal adenocarcinoma: Exploring texture features for automated classification and clinical decision support [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr LB118.