Automated COVID-19 classification in CT images using ICD u-net segmentation with CNN feature extraction and random forest classification

The COVID-19 pandemic has created an urgent need for efficient diagnostic tools to assist healthcare professionals in making timely clinical decisions. Although computed tomography scans offer valuable information on lung abnormalities, analyzing these images requires specialized expertise and is time-consuming. This study presents an automated system for COVID-19 detection that combines deep learning segmentation with machine learning classification to provide both accuracy and interpretability. The approach incorporates advanced preprocessing steps, including enhancement of the lung region through denoising and segmentation performed by an Isolated Channel-wise Denoising U-Net (ICD-U-Net), which enhances image quality and isolates relevant lung structures. Deep features are then extracted from these segmented areas using a convolutional neural network, followed by a Random Forest classification that offers transparent decision-making through feature importance analysis.The system was evaluated on a CT image data set comprising two diagnostic categories: COVID-19 and non-COVID-19.The results show that the proposed method performs better than existing approaches. Each part of the system adds to its overall accuracy, with segmentation greatly improving classification. The system provides reliable and interpretable COVID-19 detection, making it suitable for clinical use as a diagnostic tool.