3D Virtual Pancreatography.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Oncology

Background:

• Current non-invasive screening for pancreatic lesions relies on 2D CT images, often limiting visibility and automatic detection of crucial features.
• Early and accurate diagnosis of pancreatic lesions is critical for effective treatment of pancreatic cancer precursors.

Purpose of the Study:

• To introduce 3D Virtual Pancreatography (VP), a novel visualization application for non-invasive diagnosis and classification of pancreatic lesions.
• To improve upon existing methods by providing clearer visualization and automated analysis of pancreatic anatomy and pathology.

Main Methods:

• Development of an end-to-end visual diagnosis system incorporating machine learning for automatic segmentation of the pancreas and lesions.
• Utilizing a semi-automatic approach for primary pancreatic duct extraction and machine learning for lesion classification into four types.
• Integration of specialized 3D/2D visualizations, volume rendering, and pancreas/lesion-centric measurements.

Main Results:

• VP provides enhanced, end-to-end visual diagnosis capabilities for pancreatic lesions.
• The system successfully segments pancreatic structures and classifies lesions, aiding in diagnosis.
• Collaborative design with radiologists ensured clinical relevance and evaluation on real-world CT datasets.

Conclusions:

• 3D Virtual Pancreatography (VP) represents a significant advancement in the non-invasive diagnosis and classification of pancreatic lesions.
• The system's integrated approach, combining machine learning and advanced visualization, facilitates more effective detection of pancreatic cancer precursors.
• VP has been validated through expert radiologist feedback and real-world case studies, demonstrating its clinical utility.