Wizako-01/Transfer-learning-model: Explainable Transfer Learning for Multi-Class Pulmonary Disease Classification Using Nigerian Chest X-ray Dataset

This repository contains the research workflow and supporting materials for a transfer learning-based chest X-ray classification study using a local dataset from Aminu Kano Teaching Hospital, Nigeria. The project focuses on the automated classification of chest X-ray images into four categories: normal, pneumonia, coronavirus disease 2019, and tuberculosis.

The study evaluates three convolutional neural network architectures: ResNet-50, DenseNet-121, and EfficientNet-B0. The models were fine-tuned for multi-class pulmonary disease classification, and their performance was assessed using accuracy and class-level evaluation metrics. ResNet-50 achieved the highest overall accuracy of 92.67%, followed by DenseNet-121 with 91.67% and EfficientNet-B0 with 90.67%. DenseNet-121 demonstrated stronger class-level performance across multiple metrics and was further assessed using activation mapping to examine whether the model focused on clinically relevant lung regions during prediction.

This work highlights the potential of locally trained computer-aided diagnostic systems to support chest X-ray interpretation in Nigeria, where limited radiology workforce capacity and high clinical workload may contribute to diagnostic delays. By using local clinical data, the project addresses the importance of population-relevant model development and supports the safe, ethical, and context-aware integration of artificial intelligence into radiology practice.

This archive is provided for research, documentation, and reproducibility purposes. It is not intended for direct clinical diagnosis or patient management.