Deep-learned time-signal intensity pattern analysis using an autoencoder captures magnetic resonance perfusion...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Current dynamic susceptibility contrast (DSC) magnetic resonance imaging (MRI) methods struggle to fully analyze complex time-signal intensity curve dynamics.
• Accurate tissue characterization and tumor diagnosis in the brain rely on detailed analysis of dynamic MRI data.

Purpose of the Study:

• To investigate if an autoencoder-based pattern analysis of DSC MRI can capture representative temporal features.
• To determine if this approach enhances tissue characterization and improves tumor diagnosis in a multicenter setting.

Main Methods:

• An autoencoder was employed to analyze DSC MRI time-signal intensity curves, extracting representative temporal patterns.
• These patterns were then utilized by a convolutional neural network trained on 216 preoperative DSC MRI scans and validated on 43 external scans.
• Clustering identified nine distinct temporal patterns, differentiating tumor types.

Main Results:

• The autoencoder successfully identified nine representative temporal pattern clusters, accurately distinguishing tumoral from non-tumoral tissues.
• Dominant temporal pattern clusters differentiated primary central nervous system lymphoma (PCNSL) from glioblastoma (AUC 0.89) and metastasis from glioblastoma (AUC 0.95).
• The method demonstrated generalizability across different centers and acquisition protocols.

Conclusions:

• Autoencoder-based pattern analysis effectively captures complex temporal dynamics in DSC MRI.
• This approach significantly improves the identification of tumoral tissues and the differentiation between brain tumor types.
• The technique shows promise for enhanced, generalizable diagnostic capabilities in neuro-oncology.