Enhanced MobileNet Architecture with Residual Blocks for Improved CT Image Classification

This research introduces a novel modification to the MobileNet architecture, incorporating residual blocks to enhance its efficacy in the classification of computed tomography (CT) images, specifically for distinguishing malignant and benign cases. Comparative evaluations against MobileNet V1 and V2 were conducted using a diverse CT image dataset. Our proposed modified MobileNet consistently outperformed both baseline architectures across key evaluation metrics, including precision, recall, F1 score, confusion matrix, and overall accuracy. Additionally, the study investigated the impact of preprocessing techniques on classification performance. Results demonstrated a substantial improvement in accuracy when preprocessing was applied, underscoring its importance in optimizing deep learning models for medical image analysis. The incorporation of residual blocks into the MobileNet architecture proved instrumental in capturing intricate features, contributing to the observed performance boost compared to standard MobileNet architectures. This research not only highlights the significance of architectural modifications but also emphasizes the critical role of preprocessing in refining deep learning models for medical image classification. In conclusion, our findings suggest that the modified MobileNet architecture with residual blocks holds promise for enhanced CT image classification, with potential applications in early disease detection and clinical decision support systems.