Segmentation of cerebrovascular pathologies in stroke patients with spatial and shape priors

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate segmentation of cerebrovascular pathologies in brain MRIs is crucial for understanding cerebral ischemia.
• Manual delineation of lesions is time-consuming and infeasible for large-scale stroke studies.
• Lesion variability in location and shape poses segmentation challenges for traditional methods.

Purpose of the Study:

• To propose and demonstrate an automated inference algorithm for segmenting cerebrovascular pathologies in clinical brain MR images.
• To address the limitations of manual segmentation in large stroke patient cohorts.
• To develop a method capable of handling the diverse spatial patterns and intensity properties of stroke lesions.

Main Methods:

• A generative model was developed to capture spatial and intensity characteristics of cerebrovascular pathologies.
• The algorithm was trained and demonstrated on clinical MR images from a stroke patient cohort.
• The focus was on automatic segmentation, overcoming challenges posed by lesion variability.

Main Results:

• The proposed inference algorithm successfully performs automatic segmentation of cerebrovascular pathologies.
• The generative model effectively captures key pathological features in MR images.
• Demonstration on a clinical cohort validates the algorithm's applicability.

Conclusions:

• Automated segmentation of cerebrovascular pathologies is feasible and beneficial for stroke research.
• The developed algorithm offers a scalable solution for analyzing large neuroimaging datasets.
• This approach facilitates a deeper understanding of stroke mechanisms and clinical outcomes.