PFP-HOG: Pyramid and Fixed-Size Patch-Based HOG Technique for Automated Brain Abnormality Classification with MRI

Area of Science:

• Medical Imaging Analysis
• Machine Learning in Neurology
• Computational Neuroscience

Background:

• Detecting neurological abnormalities like Alzheimer's disease (AD) and brain tumors from MRI is crucial for diagnosis.
• Existing deep learning models for MRI analysis often suffer from high time complexity.
• There is a need for efficient and accurate methods to aid neurologists in screening brain abnormalities.

Purpose of the Study:

• To develop a novel, time-efficient, hand-modeled feature-based learning network for neurological abnormality detection in MRI.
• To introduce a new feature generation architecture, Pyramid and Fixed-size Patch (PFP), for enhanced classification performance.
• To reduce the time complexity associated with deep learning models while achieving high accuracy in brain abnormality detection.

Main Methods:

• Proposed a novel feature generation architecture: Pyramid and Fixed-size Patch (PFP) using handcrafted feature extractors.
• Integrated Histogram of Oriented Gradients (HOG) with the PFP architecture (PFP-HOG) for discriminative feature extraction.
• Utilized iterative Chi2 (IChi2) for feature selection and k-nearest neighbors (kNN) with tenfold cross-validation for classification.

Main Results:

• The PFP-HOG and IChi2-based model achieved high classification accuracies across multiple datasets.
• Specific accuracies include 100% for the AD dataset, 94.98% for brain tumor dataset 1, 98.19% for brain tumor dataset 2, and 97.80% for a merged dataset.
• The proposed method demonstrated robust classification performance and reduced time complexity compared to deep learning approaches.

Conclusions:

• The novel PFP-HOG feature extraction method offers an accurate and efficient approach for detecting neurological disorders from MRI.
• This model has the potential to assist neurologists by providing a reliable tool for validating manual MRI brain abnormality screening.
• The study highlights the effectiveness of handcrafted features combined with optimized selection and classification for medical image analysis.