Robust unsupervised segmentation of infarct lesion from diffusion tensor MR images using multiscale statistical...

Area of Science:

• Medical Imaging
• Neuroscience
• Biomedical Engineering

Background:

• Manual segmentation of infarction lesions in diffusion tensor magnetic resonance imaging (DT-MRI) is time-consuming in clinical practice.
• Diffusion tensor magnetic resonance imaging (DT-MRI) provides valuable data for diagnosing brain infarction but requires efficient analysis methods.

Purpose of the Study:

• To develop and validate an unsupervised, automatic method for segmenting infarction lesions in DT-MRI scans.
• To improve the efficiency and accuracy of infarction lesion detection and analysis in stroke patients.

Main Methods:

• A multistage unsupervised procedure including image preprocessing, tensor field calculation, diffusion anisotropy measurement, adaptive multiscale statistical classification (MSSC), and partial volume voxel reclassification (PVVR).
• The method addresses common DT-MR image artifacts such as noise, intensity overlapping, partial volume effect (PVE), and intensity shading.

Main Results:

• The automatic segmentation method was applied to 20 patients with diagnosed brain infarction using DT-MRI.
• Clinical experts confirmed the accuracy and reproducibility of the method in identifying infarction lesions.
• The method demonstrated effectiveness in segmenting infarction lesions and analyzing diffusion anisotropy.

Conclusions:

• The developed automatic segmentation method offers a promising, efficient alternative to manual tracing for infarction lesions in DT-MRI.
• This technique can aid in detecting lesion location and size, and quantitatively analyzing diffusion anisotropy to guide clinical diagnosis and therapy for stroke patients.