This research proposes a lightweight deep learning (LDM) approach for detecting COVID-19 in lung CT slices using a combination of Shannon's entropy thresholding and the Brownian Butterfly Algorithm (BBA) for feature optimization. The method involves image preprocessing with Shannon's entropy-based tri-level thresholding, deep feature extraction using LDMs, and feature optimization with BBA. The system then performs binary classification using three-fold cross-validation. The proposed scheme achieves high detection accuracy, with 93.80% using individual features, 96% with fused features, and 99.10% with ensemble features. The study evaluates the performance of various LDMs, including SqueezeNet, SqueezeNext, NASNetMobile, MobileNetV1, MobileNetV2, MobileNetV3_Small, and MobileNetV3_Large. The results show that the proposed method significantly improves the accuracy of COVID-19 detection in lung CT slices. The system is tested on a dataset of 10,000 2D axial-plane CT slices (5,000 healthy and 5,000 COVID-19). The study also compares the performance of different classifiers, including SoftMax, Random Forest, Decision Tree, Naive Bayes, K-Nearest Neighbour, and Support Vector Machine. The results confirm that the proposed method outperforms existing techniques in terms of detection accuracy. The study highlights the effectiveness of the proposed approach in enhancing the accuracy of COVID-19 detection in lung CT slices. The research also discusses the ethical approval, informed consent, data availability, and funding sources. The authors declare no competing interests. The study concludes that the proposed method is a promising approach for detecting COVID-19 in lung CT slices.This research proposes a lightweight deep learning (LDM) approach for detecting COVID-19 in lung CT slices using a combination of Shannon's entropy thresholding and the Brownian Butterfly Algorithm (BBA) for feature optimization. The method involves image preprocessing with Shannon's entropy-based tri-level thresholding, deep feature extraction using LDMs, and feature optimization with BBA. The system then performs binary classification using three-fold cross-validation. The proposed scheme achieves high detection accuracy, with 93.80% using individual features, 96% with fused features, and 99.10% with ensemble features. The study evaluates the performance of various LDMs, including SqueezeNet, SqueezeNext, NASNetMobile, MobileNetV1, MobileNetV2, MobileNetV3_Small, and MobileNetV3_Large. The results show that the proposed method significantly improves the accuracy of COVID-19 detection in lung CT slices. The system is tested on a dataset of 10,000 2D axial-plane CT slices (5,000 healthy and 5,000 COVID-19). The study also compares the performance of different classifiers, including SoftMax, Random Forest, Decision Tree, Naive Bayes, K-Nearest Neighbour, and Support Vector Machine. The results confirm that the proposed method outperforms existing techniques in terms of detection accuracy. The study highlights the effectiveness of the proposed approach in enhancing the accuracy of COVID-19 detection in lung CT slices. The research also discusses the ethical approval, informed consent, data availability, and funding sources. The authors declare no competing interests. The study concludes that the proposed method is a promising approach for detecting COVID-19 in lung CT slices.