Enhanced predictive accuracy of pancreatic ductal adenocarcinoma staging: A synergistic approach merging machine learning algorithms with metabolic profiling

Abstract

Early diagnosis and treatment are pivotal for enhancing the survival rates of pancreatic cancer patients, emphasizing the necessity for precise staging of pancreatic ductal adenocarcinoma (PDAC). This study presents a hybrid model that combines convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and traditional machine learning (ML) methods to predict PDAC staging based on metabolic characteristics. To address the data imbalance in PDAC datasets, the adaptive synthetic (ADASYN) sampling algorithm was utilized to augment minority class samples. The CNN-LSTM-ML hybrid model was developed and its performance was evaluated against traditional classification methods. The hybrid model achieved an optimal classification accuracy of 90.00 %, surpassing the performance of traditional methods. The confusion matrix indicated 100 % prediction accuracy for PDAC-I and PDAC-IV stages, and 66.67 % and 83.33 % for PDAC-II and PDAC-III stages, respectively. Validation across datasets with varying degrees of malnutrition confirmed the model's reliability. These results demonstrated the excellent predictive performance of the CNN-LSTM-ML hybrid model and its potential applicability to staging prediction in other clinical conditions, contributing to the advancement of precision and personalized medical interventions.