Deep learning-based end-to-end scan-type classification, pre-processing, and segmentation of clinical neuro-oncology...

Area of Science:

• Artificial Intelligence
• Neuro-oncology
• Medical Imaging

Background:

• Neuro-oncology research relies on high-dimensional MRI data, often heterogeneous due to varied acquisition protocols.
• Manual data curation and pre-processing are time-consuming and fragmented, hindering algorithmic application.
• Existing automation efforts are incomplete or require substantial manual input.

Purpose of the Study:

• To develop an end-to-end AI framework for automated processing of neuro-oncology MRI data.
• To transform raw multi-modal MRI Digital Imaging and Communications in Medicine (DICOM) data into quantitative tumor measurements.
• To streamline clinical workflows and expedite research data curation.

Main Methods:

• An AI framework was developed to classify MRI sequence types and preprocess data.
• Convolutional neural networks were employed for tumor tissue subtype segmentation.
• An expert-in-the-loop approach allowed manual refinement of segmentation results by radiologists.

Main Results:

• The scan-type classifier achieved 99.71% accuracy.
• The segmentation model attained a mean Dice Similarity Coefficient of 0.894 for whole tumor segmentation.
• The framework was validated on a retrospective glioma dataset (n=155).

Conclusions:

• The proposed AI framework automates tumor segmentation and characterization from neuro-oncology MRI data.
• This automation streamlines clinical workflows and standardizes large-scale research data curation.
• The system integrates AI-driven segmentation with expert radiologist refinement for robust results.