A lightweight model for pneumonia classification

Madagascar suffers from pneumonia, a life-threatening disease that causes several deaths around the world every year. As a developing country, Madagascar faces lack of health infrastructures and radiologists to enable the early detection of this disease. That situation contributes to the high costs that most people could not afford. Thus, deep learning methods try to surpass these limits by providing models that could assists specialists and could run in a low-resource environment like a smartphone. That reduces considerably the overall expenses and contribute to save more lives. That perspective leads this work to propose a very lightweight model inspired from the architecture ofSSDLiteX that achieves a very great performance. To get further, preprocessing was used and a combination with LiteSRGAN was tested, too. After experiments, it achieves a 100% accuracy both with thecustom preprocessing techniques and with LiteSRGAN. A validation accuracy of99.07%, that is a very great metric showing high generalization, improved to 99.69% using liteSRGAN. The model is reliable and answers the need of the health domain in Madagascar and in other developing countries.