Breast Cancer: Segmentation-Guided Hybrid Classification

Datasets: BUSI with GT (segmentation) & BUSI_Corrected (classification)

A two-stage pipeline on breast ultrasound: (1) train U-Net on BUSI-with-GT to obtain lesion masks, then segment BUSI_Corrected images;

(2) merge classes (Normal, Benign, Malignant) and train a hybrid classifier that fuses features from strong CNN backbones. evaluation metrics: Accuracy/Loss curves, ROC for all classes, confusion matrix, and a full metrics table.

Lung–Colon Histopathology: Image Enhancement & Transfer Learning

Dataset: Lung–Colon Histopathology Images

This project applied sequential image processing techniques to enhance round nuclei in lung–colon histopathological images, followed by evaluation using MSE, RMSE, SSIM, and PSNR metrics.

Five transfer learning models (ResNet50, DenseNet121, EfficientNet-B0, MobileNetV3, ViT-B/16) were exploredwith the best model by accuracy and F1-score. An ablation study has also been performed.

Image Preprocessing & Quality Analysis with Vision Transformer

Dataset: Medical Image Preprocessing Experiments

This project applied sequential preprocessing and quality enhancement steps to medical images, followed by classification using a Vision Transformer (ViT).
Performance evaluation achieved 87% testing accuracy, demonstrating the effectiveness of preprocessing in improving transformer-based models. The workflow included image quality analysis, evaluation metrics summarization.