What does this research mean for the field?

An ensemble architecture combining VGG19 and ResNet50 achieves 98.21% accuracy for COVID-19 CT scan classification, and a BAT-based optimization algorithm can effectively compress these models for deployment on resource-constrained devices without compromising accuracy or explainability. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to apply and evaluate transfer learning models for classifying COVID-19 using CT images and to use an optimization algorithm for model compression.

May 6, 2026

Pre-trained model compression using BAT Optimization Algorithm for Covid-19 images classification with XAI

Puntos clave

The study aims to apply and evaluate transfer learning models for classifying COVID-19 using CT images and to use an optimization algorithm for model compression.
Employed six pre-trained CNN models: MobileNetV2, Xception, ResNet50, ResNet152, VGG16, VGG19.
Used K5(80:20) protocol on balanced and augmented CT scan datasets.
Implemented BAT-based model compression and GRADCAM for explainability.
Accuracy of models ranged from 83.66% to 92.55%, with the ensemble model achieving 98.21%.
BAT-based compression maintained accuracy without significant loss.
GRADCAM highlighted infected areas in CT scans effectively.

Resumen

Convolutional neural networks and other deep neural networks have grown significantly in image classification in recent years. Training large deep models on a large amount of labelled data has been shown to be the best practice for image classification. However, the requirement for substantial training data to attain optimal performance is unachievable in numerous real-world situations. Transfer learning can help enhance performance in these situations. The main objective of the study is to apply and evaluate multiple transfer learning models that classify COVID-19 using CT-scan medical images and to apply an optimization algorithm to compress the model so that the model can be deployed in resource-constrained devices. High-resolution lung computed tomography (CT) methods are frequently employed in the intensive care unit (ICU) for the classification of COVID-19 disease. For the COVID-19 CT-scan binary classification task, we employed six pre-trained CNN models in this paper: MobileNetV2, Xception, ResNet50, ResNet152, VGG16, and VGG19. All models are applied to segmented CT scan data, K5(80:20) protocol is used on the balanced and augmented datasets. The accuracy of models is 83.66%, 91.37%, 85.75%, 92.55%, 86.80% and 90.98%. A new architecture was designed using an ensemble of VGG19 and ResNet50 to give best performance of 98.21%. A BAT-based model compression technique is implemented on the COVID-19 dataset to compress the pre-trained models without compromising the model accuracy. GRADCAM explainability has also been demonstrated on original and compressed models to highlight the most infected part in COVID-19 patient CT scans.

Me gusta

Guardar