Multimodal Artificial Intelligence Using Endoscopic USG, CT, and MRI to Differentiate Between Serous and Mucinous...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Serous cystic neoplasms (SCN) and mucinous cystic neoplasms (MCN) present similar imaging features, complicating diagnosis with single modalities.
• Current artificial intelligence (AI) models show promise in differentiating SCN and MCN using single-modal imaging, but diagnostic performance needs improvement.
• Multimodal imaging techniques like CT, MRI, and EUS are traditionally used for SCN and MCN differentiation.

Purpose of the Study:

• To compare the diagnostic efficacy of AI in classifying SCN and MCN using multimodal imaging versus single-modal imaging.
• To assess the performance of AI utilizing combinations of EUS, CT, and MRI for pancreatic cyst classification.

Main Methods:

• Retrospective analysis of imaging data from 25 SCN and 24 MCN patients.
• Four imaging modalities (EUS, CT, MRI, MR pancreatography) were used, with data augmentation creating 39,200 images per modality.
• A ResNet-based AI model was trained on single and combined modalities, with results compared against five experienced gastroenterologists.

Main Results:

• AI performance improved with each additional imaging modality: single (90.8% accuracy), double (94.9% accuracy), triple (97.0% accuracy), and all four (99.0% accuracy).
• AI utilizing all four modalities achieved 98.0% sensitivity and 100% specificity, significantly outperforming human experts (81.0% accuracy).
• AI models consistently surpassed expert performance across all metrics (sensitivity, specificity, accuracy).

Conclusions:

• AI leveraging multimodal imaging significantly enhances the classification accuracy of SCN and MCN.
• Multimodal AI demonstrates superior diagnostic performance compared to single-modal AI and experienced clinicians.
• The study highlights the potential of AI in improving the differential diagnosis of pancreatic cystic neoplasms.