AI Project

MediScan was developed as a research project to assist radiologists in triaging chest X-rays. The dataset used was a publicly available collection of over 5,800 labeled images. After extensive preprocessing (histogram equalization, resizing to 224×224, and data augmentation including rotation, zoom, and horizontal flip), a convolutional neural network was built using TensorFlow/Keras.

The final architecture used a pretrained ResNet50 backbone with custom dense layers, dropout for regularization, and a softmax output. Training was done on an NVIDIA RTX 3080 with early stopping and learning rate reduction. The model achieved an F1-score of 0.87 on the test set, with explainability heatmaps generated via Grad-CAM to highlight abnormal regions.

Deployment & Real-World Application

A simple Flask web app was built to allow users to upload an X-ray and receive a prediction with confidence scores. This demonstrates a complete AI pipeline: data cleaning, modeling, evaluation, interpretability, and deployment. The web interface provides an intuitive way for medical professionals to interact with the model and understand its predictions through visual explanations.

Limitations include dataset bias and the need for external validation, but the project demonstrates strong technical competency in building end-to-end machine learning solutions with practical applications in healthcare diagnostics.

The project showcases expertise in computer vision, transfer learning, model optimization, and AI deployment—proving that deep learning can be applied to solve real-world medical challenges with transparency and reliability.

NEXT

NaturaLive Platform

Responsive Web Design UX/UI Design Interactive Maps