README.md

Medical images recognition - Medical MNIST analysis

The project focuses on Medical MNIST analysis. Models for image classification are developed, the results are looked into as well as some models performance visualisations are presented.

Notebooks in the repository

• Primary notebook - dataset analysis and preparation along with model training
• Results - metrics, lerning curves, violin plots, ROC and PR curves, images from outside the dataset
• Occlusion sensitivity - occlusion sensitivity for different models
• t-SNE visualisations - mosaics, scatter plots and visualisations using RasterFairy

Examples from each notebook

Primary notebook

Average image of each class:

Results

ROC and PR curves for best model - different image sizes:

Occlusion sensitivity

Best model occlusion sensitivity for each class example:

t-SNE visualisations

colored t-SNE mosaic with images: