Department of Computer Science, COMSATS University Islamabad, Wah Campus, Wah Cantt, Pakistan, 47040.
Email: itsmetalhaimran@gmail.com.
Faculty of Computer Studies, Arab Open University, Riyadh, 11681, Saudi Arabia.
Email: s.iftikhar@arabou.edu.sa.
Technical and Further Education (TAFE), New South Wales (NSW), Australia.
Email: kiran.fatima4@tafensw.edu.au.
Faculty of Computer Studies, Arab Open University, Riyadh, 11681, Saudi Arabia.
Email: melamir@arabou.edu.sa.
Faculty of Computer Studies, Arab Open University, Riyadh, 11681, Saudi Arabia.
Email: n.alansari@arabou.edu.sa.
Department of Computer Science, COMSATS University Islamabad, Wah Campus, Wah Cantt, Pakistan, 47040.
Email: ammarsaeed1997@gmail.com.
ABSTRACT
Background: Malaria poses an enormous threat to humanity with ever increasing cases annually. The research in the field of medical is contributing quite a lot in providing methods for premature diagnosis of malaria. Apart from medical research, information technology is also playing a vital role in proposing efficient methods for malaria diagnosis. Aim: To minimize the manual interference and boost the diagnosis accuracy, the automated systems are under study lately. In the proposed work, an Optimized Deep Malaria Classifier (ODMC) is proposed for accurate and efficient malaria blood smear classification. Method: A dataset comprising of healthy and infected images of malaria blood smears is pre-processed using color space transformation and a series of other image enhancement steps. The deep features are extracted using the well-trained layers of pre-trained Convolutional Neural Networks (CNNs) including ResNet101(RSN101) and SqueezeNet (SQN). Apart from this, the local handcrafted features are also extracted from the pre-processed dataset using Local Binary Patterns (LBP). Both the deep features and the handcrafted features are serially fused together to formulate a compact feature vector which is then optimized using Linear Discriminant Analysis (LDA). The optimized vector is the classified using multiple classifier kernels. Results: The ODMC achieved 99.73% accuracy and 99.76% precision whilst maintaining an efficient prediction speed and training time.