top of page

Detecting Mitoses with a Convolutional Neural Network for MIDOG 2022 Challenge

MIDOG 2022 challenge, MICCAI 2022

Authors: Hongyan Gu, Mohammad Haeri, Shuo Ni, Christopher Kazu Williams, Neda Zarrin-Khameh, Shino Magaki, and Xiang 'Anthony' Chen

Abstract: This work presents a mitosis detection method with only one vanilla Convolutional Neural Network (CNN). Our approach consists of two steps: given an image, we first apply a CNN using a sliding window technique to extract patches that have mitoses; we then calculate each extracted patch’s class activation map to obtain the mitosis’s precise location. To increase the model gen-eralizability, we train the CNN with a series of data augmenta-tion techniques, a loss that copes with noise-labeled images, and an active learning strategy. Our approach achieved an F1 score of 0.7323 with an EfficientNet-b3 model in the preliminary test phase of the MIDOG 2022 challenge.

bottom of page