Hybrid Wavelet and K-Means Clustering Framework with Bilateral Filtering for Automated COVID-19 CT Lesion Segmentation

Abstract

Coronavirus disease (COVID-19) has emerged as a major global health challenge, making early and accurate diagnosis crucial, particularly for asymptomatic patients. Computed Tomography (CT) imaging has proven to be an effective modality for detecting COVID-19-related lung abnormalities.

In this paper, a hybrid framework integrating Wavelet Transform (WT), K-Means Clustering (K-MC), and Bilateral Filtering (BF) is proposed for automated segmentation of COVID-19 lesions in chest CT images. Initially, WT is employed for image denoising in the frequency domain to suppress noise while preserving structural details. Subsequently, K-Means clustering based on texture characteristics and local gray-level entropy is applied to achieve automatic segmentation of infected regions. The obtained segmentation results are further refined using bilateral filtering to reduce residual noise and preserve edge information. Finally, morphological post-processing operations are performed to improve segmentation accuracy and region consistency.

Experimental results obtained on a publicly available COVID-19 CT dataset demonstrate that the proposed framework achieves a mean Dice coefficient of 0.8177, a Jaccard index of 0.7322, a sensitivity of 0.8281, a precision of 0.8538, and a specificity of 0.9837. Furthermore, the proposed method outperforms four state-of-the-art segmentation approaches (CV, MAC, FRAGL, and SPF) while reducing computational time by more than 87% compared with conventional methods. These findings highlight the effectiveness and computational efficiency of the proposed framework for automated COVID-19 CT image analysis.