Medical image classifier with ResNet50

A MATLAB-based image classification project using ResNet50 to detect whether cells are affected or healthy.

Project Overview

This project focused on the development of a medical image classifier using transfer learning with ResNet50, one of the most powerful pre-trained convolutional neural networks available. The goal was to classify microscopic cell images into two categories: affected or unaffected, based on visual patterns.

It was developed in MATLAB as part of an academic initiative to explore deep learning techniques in biomedical applications.

Objectives

• Apply deep learning to a real-world medical classification problem.
• Use transfer learning to fine-tune a pretrained ResNet50 model for binary classification.
• Understand the challenges of working with limited medical image datasets.

Technologies Used

• MATLAB Deep Learning Toolbox
• ResNet50 pretrained CNN model
• Image Processing Toolbox (for preprocessing and augmentation)
• Custom dataset of cell images labeled as healthy or affected

Key Features

• Image resizing and preprocessing for ResNet50 input.
• Data augmentation to improve model generalization.
• Layer replacement and retraining of the fully connected layers.
• Real-time prediction of test images with classification accuracy output.

What I Learned

• How to implement transfer learning workflows in MATLAB.
• Techniques for managing class imbalance and limited datasets.
• Evaluation using confusión matrices and accuracy metrics.
• The potential and limitations of AI in medical diagnostics.

Final Thoughts

This project gave me valuable experience in the application of deep learning to medical image analysis. It emphasized the importance of data quality, model interpretability, and performance validation -- especially when the stakes involve clinical decisions.