Energy-Efficient Analog-Compression-Driven Imaging for Edge Devices in IoT Healthcare Systems

This paper presents an analog compression and low-bit-depth quantization architecture designed for low power medical imaging systems operating within IoT and edge-AI environments. The proposed front-end performs 8 × and 64× analog compression along with 2.5-bit quantization within a single conversion cycle, significantly reducing data volume before digitization. This results in substantial savings in ADC power and bandwidth, enabling low-energy transmission for IoT-connected healthcare devices. The conventional CNNs trained on original high-resolution datasets (ChestXray2017 and Brain MRI) retain > 90% classification accuracy even when tested directly on compressed, low-bit-depth images generated by the proposed circuit— without any model retraining, fine-tuning, or curation for low-resolution data.