Enhancing Surgical Planning with AI-Driven Segmentation and Classification of Oncological MRI Scans

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Anatomy

Background:

• Current medical image processing pipelines require optimization for patient-specific analysis.
• Artificial Intelligence (AI) offers potential for enhanced accuracy and efficiency in processing oncological Magnetic Resonance Imaging (MRI).

Purpose of the Study:

• To develop an AI-based pipeline for patient-specific 3D reconstruction from oncological MRI.
• To improve the accuracy and efficiency of medical image processing for enhanced clinical utility.

Main Methods:

• Development of an AI pipeline integrating automatic MRI sequence classification (ResNet) and anatomical segmentation (nnU-Net v2).
• Leveraging sequence-aware information for improved understanding of MRI signals and enhanced delineation of anatomical structures.

Main Results:

• Achieved over 90% accuracy in MRI sequence classification.
• Demonstrated improved segmentation performance, particularly for contrast-sensitive anatomies (hepatic vasculature, pancreas) and musculoskeletal structures.
• Complete processing time for an MRI case was approximately four minutes.

Conclusions:

• The AI pipeline provides accurate patient-specific 3D reconstructions from oncological MRI.
• Integration of sequence-aware information enhances delineation accuracy, outperforming existing methods.
• The developed system shows potential for integration into surgical planning workflows, advancing AI-driven medical image analysis.