Machine Learning-Based Detection of Dengue from Blood Smear Images Utilizing Platelet and Lymphocyte Characteristics

• DOI: 10.3390/diagnostics13020220
• Publication Date: 2023-01-09T08:25:58Z
• Authors: Hilda Mayrose; G. Muralidhar Bairy; Niranjana Sampathila; Sushma Belurkar; Kavitha Saravu
• Keywords: Medicine (General); digital pathology; dengue; machine learning; thrombocytopenia; lymphocyte; peripheral blood smear; Artificial Intelligence; thrombocytopenia; 02 engineering and technology; lymphocyte; dengue; Article; 3. Good health; machine learning; R5-920; peripheral blood smear; 0202 electrical engineering, electronic engineering, information engineering; digital pathology

Dengue fever, also known as break-bone can be life-threatening. Caused by DENV, an RNA virus from the Flaviviridae family, dengue is currently a globally important public health problem. The clinical methods available for diagnosis require skilled supervision. They are manual, time-consuming, labor-intensive, and not affordable to common people. This paper describes method that support clinicians during diagnosis. It proposed automate peripheral blood smear (PBS) examination using Artificial Intelligence (AI) aid Nowadays, AI, especially Machine Learning (ML), increasingly being explored successful analyses in biomedical field. Digital pathology coupled with AI holds great potential developing healthcare services. automation system developed incorporates blob detection detect platelets thrombocytopenia PBS images. results achieved clinically acceptable. Moreover, ML-based technique images of based on lymphocyte nucleus. Ten features extracted, including six morphological four Gray Level Spatial Dependance Matrix (GLSDM) features, out nucleus normal cases. Features then subjected various popular supervised classifiers built ten-fold cross-validation policy automated detection. Among all classifiers, best performance was Support Vector (SVM) Decision Tree (DT), each accuracy 93.62%. Furthermore, 1000 deep extracted pre-trained MobileNetV2 177 textural Local binary pattern (LBP) feature selection. ReliefF selected 100 most significant fed classifiers. attained SVM classifier 95.74% accuracy. With obtained results, it evident this approach efficiently contribute adjuvant tool diagnosing digital microscopic PBS.