Energy-Efficient XceptionNet Accelerator with E-FPGAs for XAI-Based Risk Prediction of Skin Abnormalities: Lecture Notes in Networks and Systems

The present era has witnessed an increase of skin cancer incidences across the globe. This can initially be observed as small patches on the human skin. Confirmation of the diease via traditional clinical tests is not only time-consuming but also an economic burden for the patient. Deep learning technology can be used to classify photographic images of such patches as low-risk infections and high-risk infections. Based on this classification, proper medication can be provided. In this work, we initially develop an intelligent model with XceptionNet to predict the risk of skin abnormalities. However, XceptionNet model is resource intensive and significant overhead in time and power is incurred that makes the system energy inefficient. We analyze the model and order the intricate functions based on their latency and power consumption. Based on the availability of embedded field programmable gate arrays (e-FPGAs), we export the functionalities into reconfigurable hardware or FPGAs that accelerate its working and make it more energy efficient. Additionally, to gain confidence of the patients and enhance transparency, we explain the working of the model with Grad-CAM visualization, where the regions in the images are highlighted, based on which the decision is taken. Experimental results on standard photographic images of skin abnormalities depict the prospects of our proposed model. Additionally, we depict the usefulness of using e-FPGAs in the design via increase in throughput and decrease in power consumption with the availability of e-FPGA resources in the system. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.