CNN Based Model for Malaria Diagnosis with Knowledge Distillation
K. M. Faizullah Fuhad, Jannat Ferdousey Tuba, Tanzilur Rahman, Nabeel Mohammed · 2020
Malaria is a deadly disease caused by plasmodium parasites and carried by female anopheles mosquitoes. Bangladesh is one of the major malaria prone country with poor infrastructures, unable to diagnose Malaria rapid at a low cost. Here we propose a custom8 layers CNN architecture that is only 233.60KB in size meaning it can easily fit and work into a low cost smartphone. The model has been optimized through different preprocessing methods and model pruning techniques like knowledge distillation ensuring that the accuracy does not reduce due to smaller model size. The accuracy received from the optimized model is~ 96.51% when tested on NIH data-set containing images of infected and uninfected cells. Proposed model is a big step towards a complete mobile based rapid diagnostic platform that can be used in any resource restricted and hilly areas of Bangladesh.