Multimodal AI approach combining deep learning imaging and clinical machine learning for pancreatic cancer detection

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Pancreatic ductal adenocarcinoma (PDAC) diagnosis is challenged by late detection, CT imaging limitations, and inaccurate biomarkers, leading to poor prognosis.
• Current diagnostic methods require improvement for timely and accurate identification of PDAC.

Purpose of the Study:

• To develop and validate a multimodal artificial intelligence (AI) approach integrating deep learning (DL) for imaging analysis and machine learning (ML) for clinical data to enhance PDAC diagnosis.
• To improve the sensitivity, specificity, and overall accuracy of pancreatic cancer detection.

Main Methods:

• A retrospective cohort of 158 patients (123 PDAC, 35 benign) was analyzed.
• A YOLOv8-based DL model was trained on contrast-enhanced CT scans for lesion detection.
• A Random Forest ML classifier analyzed clinical data (age, sex, CA19-9).
• A multimodal fusion model combined predictions from DL and ML models.

Main Results:

• The DL imaging model achieved high tumor detection performance (mAP: 87.0%, precision: 86.5%, recall: 81.2%).
• The ML clinical model demonstrated excellent specificity (precision: 100%, ROC-AUC: 0.931) but limited sensitivity (60%).
• The multimodal AI fusion model significantly outperformed individual models in sensitivity, specificity, and overall diagnostic accuracy.

Conclusions:

• A multimodal AI strategy integrating DL imaging analysis and ML clinical predictions markedly enhances diagnostic performance in pancreatic cancer.
• This AI approach shows potential as an effective decision-support tool for earlier diagnosis and optimized clinical decision-making.