Compressed Deep Learning Models with XAI for COVID-19 Detection Using CXR Images

Significant global disruptions brought on by the COVID-19 pandemic call for fast and accurate detection techniques to stem the disease’s spread. Due to its non-invasiveness and reasonable cost, chest X-ray (CXR) imaging has become a useful technique for diagnosing COVID-19. The analysis of CXR images using compressed deep learning models combined with explainable artificial intelligence (XAI) is a novel method for COVID-19 identification presented in this paper. To lower the computational complexity and memory needs of the deep learning models without compromising performance, our suggested approach uses model compression techniques including pruning and quantization. The incorporation of XAI increases transparency and makes it easier to identify key features for COVID-19 detection in CXR pictures by giving insights into the models’ decision-making process. On a sizable dataset of CXR images, we test our method, and we show that it is effective in obtaining high detection accuracy while keeping a small model size and low computing cost. Our study contribute the creation of effective, open, and trustworthy COVID-19 detection tools, which can be particularly helpful in resource-constrained environments and for enhancing confidence in AI-driven diagnostics.