A Multi-Modal Decision Support System for Automated Malaria Severity Level Assessment

The increasing demand for improved malarial control methods is directly related to its significance as one of the major health concerns of sub-Saharan African region, where accurate severity level detection is very much pertinent due to frequent improper treatment leading to complications. Many existing diagnosis methods used can classify only two types (positive or negative); however it has limitation for further decision for management for the clinical outcome which is based on three classes. This study proposed a novel multi modal clinical decision support system (CDSS) to achieve automatic malaria severity classification. Using Hybrid OOSAD + CRISP-DM approach, a CNN for image analysis (blood smear slide analysis) coupled with a RF for symptoms analysis has been designed. Decision fusion layer calculates combined parasitemia density from image with symptoms scoring, thereby resulting into classification of 3 levels of malaria viz. none, uncomplicated and severe malarial status. The performance has been evaluated and result indicate that proposed hybrid approach (CNN + RF fusion) performed best and gained maximal accuracy 97.20%, AUC-ROC 0.981 and also Excellent score for system usability scale (82.5). It has overcome basic malaria detection with actual severity classification and is a more trustworthy, swift and efficient and a scalable solution to assist health professionals for rapid decision in endemic resource constrained area and also help to minimize global malarial mortality.