Samridhi2802/Sickle-Cell-Anemia-Detection

AI model for sickle cell anemia diagnosis using combined numerical & image data. Transfer learning with EfficientNet-B0 (images) & SVM (numerical) achieved best accuracy. LIME & Grad-CAM used for explainability, enabling faster, accurate, and accessible detection in low-resource settings. # An Effective Multimodal Framework for Sickle Cell Anemia Detection using Transfer Learning and Explainable AI --- ## 🩸 Overview This project presents a multimodal, explainable AI framework for detecting **Sickle Cell Anemia (SCA)** using both **clinical blood report data** and **blood smear images**. It leverages **transfer learning** (EfficientNet-B0), **Support Vector Machine (SVM)** classifiers, and **Explainable AI** tools like **Grad-CAM** and **LIME** to provide transparent and accurate diagnosis. > 🔍 The framework is designed for clinicians and researchers to achieve faster, more interpretable, and scalable diagnosis of SCA—especially in low-resource settings. --- --- ## 🎯 Objectives * ✅ Apply **image augmentation** techniques to improve model generalization. * ✅ Develop multimodal models: * Image-based CNN using **EfficientNet-B0** * Text-based SVM for hematological data * ✅ Integrate **XAI methods**: Grad-CAM & LIME * ✅ Create an interactive **Streamlit-based UI** supporting: * Text-only input * Image-only input * Combined (multimodal) input --- ## 🧠 Methodology ### 🔬 1. Image-Based Pipeline * **Dataset**: * **AneRBC Dataset**: 12,000 RBC images * **Sickle-Specific Dataset**: 422 positive, 147 negative * **Transfer Learning Models**: * ResNet-18, DenseNet-121, EfficientNet-B0, Custom CNN * **Best Performer**: `EfficientNet-B0` (99.1% accuracy) * **Explainability**: `Grad-CAM` visualizations overlaid on input images ### 📊 2. Text-Based Pipeline * **Clinical Features**: * RBC, PCV, MCV, MCH, MCHC, RDW, TLC, Platelet Count, HGB * **Model**: `SVM` * With **SMOTE** oversampling for class balance * Achieved 96% accuracy * **Explainability**: `LIME`-based feature attribution per prediction ### ⚙️ 3. Multimodal Ensemble * Combines image and text predictions via **soft voting** * Provides **combined diagnostic result** * UI supports all 3 modes (Text, Image, Both) --- ## 🖥️ User Interface (Streamlit) | Input Type | Models Used …