A Multimodal AI Framework for the Early Identification of Pancreatic PaNET Cancer: Combining Biomarkers, Genetic Predisposition, and Radiological Data

Pancreatic neuroendocrine tumors (PanNETs) are rare yet clinically significant neoplasms characterized by heterogeneous biological behavior and often late presentation, posing substantial diagnostic challenges. Despite advancements in imaging and molecular diagnostics, current modalities exhibit limited sensitivity and specificity, particularly in early-stage or non-functional tumors. This study introduces PaNET-MultiDx, a novel multimodal artificial intelligence (AI) framework designed to improve early detection and stratification of PanNETs by integrating radiologic, genomic, and serologic data. The model architecture comprises a convolutional neural network (CNN) trained on contrast-enhanced CT and MRI scans, a multilayer perceptron for genomic mutations (e.g., MEN۱, DAXX, ATRX), and a random forest classifier for serologic biomarkers including chromogranin A and neuron-specific enolase. Outputs from these modalities are fused using an attention-based integration layer to capture cross-modal interactions. Interpretability is facilitated through Grad-CAM visualization of imaging data. Preliminary results demonstrate diagnostic accuracies exceeding ۹۰% across individual and integrated modalities, with superior performance in subtype classification, tumor grading, and early lesion detection. Notably, the model enhances sensitivity in asymptomatic and non-functional cases. PaNET-MultiDx offers a promising tool for precision diagnostics, warranting further validation in multicenter clinical settings to establish its generalizability and clinical impact.

کلیدواژه ها:

Pancreatic neuroendocrine tumors ، adiogenomics ، CNN ، PaNET-MultiDx