/OxML-Carcinoma-CLF

This project was part of the MLx Cases section of the OxML 2023 summer school.

Primary LanguageJupyter NotebookMIT LicenseMIT

Carcinoma Classification - OxML 2023 MLx Cases

Competition info

In this competition, the goal was to classify HES stained histopathological slices as containing or not containing carcinoma cells. If a carcinoma is present, it is also possible to tell whether it is malignant or not, which gives us three classes:

Carcinoma neg (-1) Carcinoma pos, benign (0) Carcinoma pos, malignant (1)

A total of 186 images was provided but labels were only present for 62 of them.

Models - Results

Two different architectures were used. The first is an Inception ResNet v2 CNN with the bottom layers frozen and the second was a zero-shot model from Open Clip. Both models achieved a score of 77.419% on the public test set and 70.967% on the private test set.