UNISELF: A Unified Network with Instance Normalization and Self-Ensembled Lesion Fusion for Multiple Sclerosis Lesion...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Manual segmentation of multiple sclerosis (MS) lesions from MRI is time-consuming and prone to variability.
• Deep learning (DL) methods offer improved efficiency and reproducibility but struggle with in-domain accuracy and out-of-domain generalization.
• Existing DL models trained on limited single-source data often show unsatisfactory performance on diverse datasets.

Purpose of the Study:

• To develop a novel DL method, UNISELF, for accurate MS lesion segmentation.
• To enhance both in-domain accuracy and out-of-domain generalization of MS lesion segmentation models.
• To address challenges posed by domain shifts and missing contrasts in multicontrast MR images.

Main Methods:

• UNISELF utilizes test-time self-ensembled lesion fusion for improved segmentation accuracy.
• Test-time instance normalization (TTIN) of latent features is employed to handle domain shifts and missing contrasts.
• The method was trained on the ISBI 2015 longitudinal MS segmentation challenge dataset.

Main Results:

• UNISELF achieved top performance on the ISBI 2015 challenge test dataset.
• The method demonstrated superior performance compared to benchmarks on diverse out-of-domain datasets (MICCAI 2016, UMCL, private multisite).
• UNISELF effectively handled domain shifts and missing contrasts caused by variations in acquisition protocols and scanner types.

Conclusions:

• UNISELF offers a robust solution for automated MS lesion segmentation, balancing in-domain accuracy and out-of-domain generalization.
• The proposed method shows significant potential for clinical applications requiring reliable and reproducible lesion quantification.
• UNISELF provides a valuable tool for MS research by enabling consistent analysis across varied imaging data.