CNN-based Diagnosis from Medical Imaging: Leveraging Transfer Learning for Enhanced Accuracy

Abstract

Medical imaging plays a crucial role in disease diagnosis and treatment planning, yet the increasing volume of imaging data poses challenges for radiologists and healthcare systems. This study investigates the application of Convolutional Neural Networks (CNNs) for automated medical image classification, leveraging transfer learning to enhance diagnostic accuracy. Using a pre-trained DenseNet-121 model, we developed and evaluated CNN-based classifiers for lung cancer, pneumonia, and tuberculosis detection. Our models achieved high classification accuracy, with 90.48% for lung cancer, 91.83% for pneumonia, and 99.84% for tuberculosis, demonstrating their effectiveness in distinguishing between normal and pathological cases. The results highlight the potential of AI-driven diagnostics to assist medical professionals by reducing workload, improving diagnostic speed and accuracy, and addressing the shortage of radiologists. Despite promising performance, challenges such as dataset variability, potential biases, and real-world deployment remain. Future work will focus on expanding datasets, improving model interpretability, and integrating AI-assisted tools into clinical workflows to enhance reliability and accessibility in medical imaging.