A Hybrid Approach for Sub-Acute Ischemic Stroke Lesion Segmentation Using Random Decision Forest and Gravitational...

Area of Science:

• Medical Imaging
• Machine Learning
• Neurology

Background:

• Sub-acute ischemic stroke is a leading global cause of death.
• Accurate segmentation of brain lesions is crucial for understanding stroke impact.
• Existing methods struggle to differentiate stroke from other brain pathologies.

Purpose of the Study:

• To develop and evaluate a novel hybrid technique for segmenting ischemic stroke lesions in Magnetic Resonance (MR) images.
• To accurately differentiate ischemic stroke from other neurological conditions like Multiple Sclerosis (MS).

Main Methods:

• A hybrid approach combining Random Decision Forest (RDF) and Gravitational Search Algorithm (GSA) was utilized.
• RDF, a machine learning algorithm, was optimized using GSA for parameter tuning (number of trees, leaves per tree).
• The GSA-RDF classifier was applied to segment ischemic stroke lesions in MR images.

Main Results:

• The proposed GSA-RDF technique achieved Root Mean Square Error (RMSE) of 16.5485%, Mean Absolute Percentage Error (MAPE) of 7.2654%, and Mean Bias Error (MBE) of 2.4585%.
• The algorithm demonstrated efficient processing of large datasets.
• Experimental results indicate superior precision and performance compared to existing segmentation methods.

Conclusions:

• The hybrid GSA-RDF classifier offers a promising new method for accurate ischemic stroke lesion segmentation in MR images.
• This technique shows enhanced precision and execution, outperforming current approaches.
• The study highlights the potential of hybrid machine learning algorithms in neuroimaging analysis.