Optimised lightweight network framework: accurate detection of ground-glass opacity and consolidation in medical computed tomography images

Accurate lung opacification diagnosis using computed tomography scans is vital for infection detection across ground-glass opacities and consolidation regions. This research proposes a novel Adaptive Particle Swarm optimised Lightweight Generative Adversarial based Deep Belief Network for the accurate detection and classification of Ground-glass opacities and consolidation regions using computed tomography images. A Lightweight Attention-based Generator-Discriminator is introduced to improve the diversity of the generated samples and acquire high-resolution synthesised computed tomography scans through the use of a Lightweight Generative Adversarial Network, incorporating the Lightweight Multi-Head Attention mechanism. The Adaptive Segmentation module is utilised to perform lesion boundary segmentation on ground-glass opacities and consolidation regions. A Deep Belief Network is adopted for the classification of the lung computed tomography scans into appropriate classes of opacification based on the features derived from the lesion regions. Severity analysis is performed by analysing the percentage involvement of ground-glass opacities and consolidation with respect to the overall lung volume. An Adaptive Particle Swarm Optimisation is utilised with a varying inertia factor for effective searching of the solution space. Extensive evaluation on different datasets using various parameters indicates that the proposed technique attains a higher accuracy of 98.93%, which is superior to existing techniques.