Deep learning-based MVIT-MLKA model for accurate classification of pancreatic lesions: a multicenter retrospective...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate differentiation of pancreatic lesions is crucial for patient management.
• Computed tomography (CT) is a key imaging modality for pancreatic lesions.
• Deep learning offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and validate a novel deep learning network for classifying pancreatic lesions using baseline CT images.
• To assess the performance of the developed model against traditional and advanced methods.
• To evaluate the impact of the model on radiologists' diagnostic performance.

Main Methods:

• A retrospective study of 864 patients with histopathological confirmation.
• Development of a hybrid deep learning model (MVIT-MLKA) integrating CNN and Transformer architectures.
• Internal and external validation of the model's performance, including comparison with other AI models and radiologist performance.

Main Results:

• The MVIT-MLKA model achieved high AUC values (0.974 training, 0.935 internal, 0.924 external).
• The model outperformed traditional machine learning and other deep learning models.
• Radiologists using the MVIT-MLKA model showed significantly improved diagnostic accuracy and sensitivity.

Conclusions:

• The MVIT-MLKA model effectively differentiates benign from malignant pancreatic lesions.
• This AI tool has the potential to reduce diagnostic errors and optimize treatment strategies.
• Integration into clinical practice can enhance pancreatic lesion classification and patient care.