Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Impact of imperfect annotations on CNN training and performance for instance segmentation and classification in digital pathology

Laura Gálvez Jiménez, Christine Decaestecker

TL;DR

This work investigates the conditions for determining an appropriate number of training epochs to prevent overfitting to annotation noise during training and indicates that the utilisation of a small, correctly annotated validation set is instrumental in avoiding overfitting and maintaining model performance to a large extent.

Abstract

Segmentation and classification of large numbers of instances, such as cell nuclei, are crucial tasks in digital pathology for accurate diagnosis. However, the availability of high-quality datasets for deep learning methods is often limited due to the complexity of the annotation process. In this work, we investigate the impact of noisy annotations on the training and performance of a state-of-the-art CNN model for the combined task of detecting, segmenting and classifying nuclei in histopathology images. In this context, we investigate the conditions for determining an appropriate number of training epochs to prevent overfitting to annotation noise during training. Our results indicate that the utilisation of a small, correctly annotated validation set is instrumental in avoiding overfitting and maintaining model performance to a large extent. Additionally, our findings underscore the beneficial role of pre-training.

Impact of imperfect annotations on CNN training and performance for instance segmentation and classification in digital pathology

TL;DR

This work investigates the conditions for determining an appropriate number of training epochs to prevent overfitting to annotation noise during training and indicates that the utilisation of a small, correctly annotated validation set is instrumental in avoiding overfitting and maintaining model performance to a large extent.

Abstract

Segmentation and classification of large numbers of instances, such as cell nuclei, are crucial tasks in digital pathology for accurate diagnosis. However, the availability of high-quality datasets for deep learning methods is often limited due to the complexity of the annotation process. In this work, we investigate the impact of noisy annotations on the training and performance of a state-of-the-art CNN model for the combined task of detecting, segmenting and classifying nuclei in histopathology images. In this context, we investigate the conditions for determining an appropriate number of training epochs to prevent overfitting to annotation noise during training. Our results indicate that the utilisation of a small, correctly annotated validation set is instrumental in avoiding overfitting and maintaining model performance to a large extent. Additionally, our findings underscore the beneficial role of pre-training.

Paper Structure

This paper contains 29 sections, 1 equation, 3 figures, 14 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related works
  3. Impact of imperfect annotations
  4. Approaches to avoid overfitting to annotation noise during training
  5. Material and methods
  6. Datasets
  7. Description
  8. Preprocessing: tiling, augmentation and class balancing
  9. Annotation corruption in clean training sets
  10. Model and training process
  11. Training stopping in the presence of noisy datasets
  12. Performance metrics
  13. Experiments and results
  14. Impact of annotation noise on the ability to stop training appropriately
  15. Impact of detection noise on Hover-Net performance
  16. ...and 14 more sections

Figures (3)

  • Figure 1: (a-c) Examples of annotations and (d-f) their corruption introduced in the clean MoNuSAC training subset. The classification masks show epithelial cells in red, lymphocytes in blue and neutrophils in green.
  • Figure 2: Loss obtained when evaluating Algorithm \ref{['hover_training']} training stages
  • Figure A.1: Flowchart of the experiments done (see details in the main text).