Refining PD classification through ensemble bionic machine learning architecture with adaptive threshold based image denoising

• Published in: Biomedical signal processing and control Vol. 85; p. 104870
• Main Authors: Redhya, M., Sathesh Kumar, K.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.08.2023
• Subjects: Ebola optimiser; Ensemble and classifier; Filtering; MR image; Parkinson disease; Random forest; Salt and pepper noise; SMA; SSA; SVM; Threshold; Weighted function; XGBoost; SSA; Weighted function; Filtering; SMA; Parkinson disease; Random forest; SVM; MR image; XGBoost; Ebola optimiser; Ensemble and classifier; Threshold; Salt and pepper noise
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2023.104870

•In this work, ensemble bionic machine learning model is proposed for the classification of Parkinson’s disease (PD) from healthy controls using MR images.•Adaptive median filter along with threshold computation is used for picture denoising. Image enhancement factor (IEF), peak signal-to-noise ratio (PSNR) and Mean square error (MSE) are assessed to objectively measure the denoising effect of the given model.•The first and second order statistical features are extracted in the feature extraction stage,•Adapted fused slime salp mould (AFSSM) soft computing approach is used in feature selection.•Ebola optimized ensemble SVM, XGBoost, and Random Forest machine learning model is used for PD detection. Parkinson's disease (PD) manifests as a loss of dopamine-producing cells present in the substantia nigra region of the brain’s central nervous system (CNS). The proposed model influences magnetic resonance imaging (MRI) for detection. The research paper consist of four stages: pre-processing, feature extraction, feature selection and classification. The proposed model influences MR images for detection during the pre-processing stage, composed with an adaptive median filter along with threshold computation. The primarily getting a binary image is that it reduces the complexity of the data and makes the recognition and classification processes easier. To reduce the training period, reducing over fitting and enhancing the accuracy, first and second order features are extracted. In third stage, adapted fused slime salp mould (AFSSM) soft computing approach is used in feature selection. The AFSSM give a fixed threshold for the selection process and find optimum values very rapidly. The ensemble machine learning model (SVM, Random Forest and XGBoost) is introduced along with Ebola optimisation through voting method. The proposed achieves the mean square error, peak signal-to-noise ratio and image enhancement factor values of 4.15, 39.1 and 0.92 respectively when compared with the existing filtering models like mean, Kalman, gaussian and weiner filter. Along with this, the performance measures like accuracy, sensitivity, precision, F1-score and specificity are examined and attains an outcome of 98.4%, 98.4%, 98%, 98.8% and 98.2%, respectively.