HEPTECH AIME ML&VA on Clouds

Detecting nuclei with deep learning and image style transfer

Not scheduled

15m

Mátyás Hall (Groundfloor) (Hotel Mercure Budapest)

Poster

Speaker

Réka Hollandi (Hungarian Academy of Sciences, Biological Research Center)

Description

A novel method for nuclei detection is proposed to process diverse microscopy images. Our method incorporates deep learning techniques such as automatic training data generation from input test images called image style transfer learning that allows adjustment to the test set prior to training even with limited data size. The proposed method was originally designed for the Kaggle Data Science Bowl 2018 competition and currently outperforms the winner solution.