Adaptive Convolutional Neural Network Integrating Sequential Memory and Transfer Learning for CT-Based COVID-19 Detection

COVID-19, first identified in Wuhan, China in 2019, is a highly contagious respiratory disease with symptoms such as fever, dry cough, and shortness of breath. Computed tomography (CT) scans are a key tool for detecting lung abnormalities related to COVID-19. However, existing approaches to COVID-19 diagnosis often struggle to extract clinically relevant features from CT images, particularly when there is inter-slice variability or limited annotated data. In this study, we introduce an Adaptive Convolutional Neural Network (ACNN) model designed to address these challenges by integrating two core mechanisms: (1) a sequential memory component using Long Short-Term Memory (LSTM) units to capture contextual relationships and dependencies across consecutive CT slices, and (2) transfer learning—leveraging pre-trained weights from large-scale medical imaging datasets to improve feature generalization. This adaptive design differs from standard CNN architectures by explicitly modeling both spatial and limited sequential information in CT scan volumes. The ACNN was trained and evaluated on the SARS-CoV-2 CT dataset, and its performance was assessed using standard metrics. Experimental results show that ACNN outperforms classical machine learning algorithms (such as KNN and SVM) and established deep learning models (including VGG16, ResNet, and DenseNet), achieving an accuracy of 97.5%, a precision of 97.30%, a recall of 97.85%, and an F1-score of 97.58%. Statistical tests confirmed the robustness of these improvements. The results demonstrate that the proposed ACNN, through its memory-augmented and transfer learning-driven design, offers a precise and reliable approach for COVID-19 diagnosis and holds promise for real-world clinical applications.