NOVA: A Novel Multi-Scale Adaptive Vision Architecture for Accurate and Efficient Automated Diagnosis of Malaria Using Microscopic Blood Smear Images

다국어 초록 (Multilingual Abstract)

NOVA: A Novel Multi-Scale Adaptive Vision Architecture for Accurate and Efficient Automated Diagnosis of Malaria Using Microscopic Blood Smear Images Md Nayeem Hosen (Advisor: Prof, Hee-Cheol Kim, PhD) Department of Computer Engineering Graduate School of Inje University Malaria continues to be a significant global health concern, particularly in tropical and subtropical areas. Although conventional approaches for diagnosis are effective, they have certain accuracy, time-consuming, and manual labor constraints. In this paper, we introduce NOVA (Novel Multi-Scale Adaptive Vision Architecture) architecture, a multi-scale adaptive vision architecture, for the diagnosis of malaria. NOVA is based on an innovative dynamic channel attention and learnable temperature spatial pyramid attention to achieve more powerful feature representation and better classification performance. In addition, adaptive feature refinement and enhanced transformer blocks are used to obtain multi-scale feature extraction and contextual reasoning. Furthermore, a multi-strategy pooling mechanism that fuses average, max, and attention-based aggregation is developed to enhance the model’s discriminative capability. We conduct experiments on a publicly accessible dataset of 15031 microscopic thin blood smear images to validate the effectiveness of the proposed approach. The model is assessed and compared on a benchmark malaria microscopy dataset, achieving an accuracy of 97.00%, precision of 96.00% and F1-score of 97.00% outperforming other existing models. The results indicate the potential of the proposed approach as a reliable automated malaria diagnostic tool. This research highlights the promise of AI-driven technologies in improving malaria diagnostics and lays the groundwork for further applications in related blood-borne illnesses.

목차 (Table of Contents)

• Ⅰ. Introduction. 10
• 1.1 Background 10
• 1.2 Motivation and Objectives 12
• 1.3 Thesis Organization 14
• Ⅱ. Literature Review. 15
• Ⅲ. Materials and Methods 18
• 3.1 Dataset Collection 18
• 3.2 Preprocessing and data splitting 19
• 3.3 Backbone Network. 22
• 3.4 Dynamic Channel Attention Module 22
• 3.5 Spatial Pyramid Attention Module 24
• 3.6 Adaptive Feature Refinement Module 25
• 3.7 Feature Enhancement Module 26
• 3.8 Enhanced Transformer Block 27
• 3.9 Multi-Strategy Pooling 28
• 3.10 Classification Head 30
• 3.11 Performance Measures 30
• IV. Models and Methods. 34
• 4.1 Methodology. 34
• 4.2 Proposed Novel NOVA Architecture. 35
• V. Results 36
• 5.1 Experimental Setup 36
• 5.2 Ablation Study 36
• 5.3. Comparison with Dataset-2 38
• VI. Discussion 44
• VII. Limitation and Future Work 45
• VIII. Conclusion 46
• IX. References 47