A Comprehensive Review of MobileNet Architectures in Medical Imaging

The digital transformation of healthcare has intensified the demand for deep learning models that are not only accurate but also computationally efficient. While heavy-weight architectures achieve state-of-the-art results in resource-rich environments, their deployment in Point-of-Care (POC) diagnostics, Internet of Medical Things (IoMT), and mobile health remains a significant challenge. The MobileNet family, characterized by depthwise separable convolutions, has revolutionized lightweight neural network design. This paper provides a systematic review of the evolution of MobileNet from its inception to the latest Neural Architecture Search (NAS)-guided MobileNetV۳ and the ۲۰۲۴-۲۰۲۵ Universal Inverted Bottleneck (UIB) innovations of MobileNetV۴ within the context of medical imaging. We analyze the theoretical mechanisms of these models and evaluate their performance across diverse modalities, including COVID-۱۹ detection, brain tumor classification, and histopathology. Beyond technical metrics, we address critical socio-technical aspects such as Explainable AI (XAI), adversarial robustness, and ethical bias. Finally, we propose a unified framework for integrating lightweight models into clinical decision support systems (CDSS) and discuss future trajectories involving Federated Learning and medical foundation models.

کلیدواژه ها:

MobileNet ، Medical Imaging ، Deep Learning ، Explainable AI (XAI) ، IOMT