Chest X-ray AI — disease detection & report generation with explainable AI

About the project

This thesis builds an AI assistant for chest X-rays. The model sorts each X-ray into one of 15 classes, the 14 conditions of the NIH ChestX-ray14 label set plus "No Finding", such as pneumonia, edema, nodules and an enlarged heart. Explainable AI shows which parts of the image led to the result, and the system turns its findings into a short written report.

On the test set it reached 99.72% accuracy (weighted F1 99.72%, precision 99.73%), and every class scored an ROC-AUC above 0.99. The student received the code, a 54-page LaTeX thesis, presentation slides and a research paper, with AI-detection and plagiarism scores under 20%.

Key features

• Multimodal deep learning

  Image and text models work together to detect disease and describe it.

• 15-class detection

  14 chest conditions such as pneumonia, edema and nodules, plus "No Finding".

• Explainable AI

  Highlights the parts of the X-ray behind each result, so doctors can check it.

• Automatic report

  Turns the findings into a short, readable report.

• Full evaluation

  Accuracy, F1, precision, recall, ROC-AUC per class and a confusion matrix.

• Thesis-ready write-up

  54-page LaTeX thesis, slides and paper, under 20% AI and plagiarism.

How it works

1. 1Upload chest X-ray
3. 2Clean & resize image
5. 3Deep model finds disease
7. 4Explainable AI shows why
9. 5Report generated

Tech stack

Deep learning

PythonDeep learningMultimodal modelExplainable AI

Evaluation

scikit-learnROC-AUCConfusion matrixGoogle Colab

Write-up

LaTeXOverleafSlidesResearch paper

What you get with a project like this

• Colab notebooks & full source code
• Trained model, ready to use
• Result charts & explainable AI images
• LaTeX thesis in your university format
• Research paper & presentation slides
• Viva prep & one-to-one walkthrough