Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ensemble of ConvNeXt V2 and MaxViT for Long-Tailed CXR Classification with View-Based Aggregation

Yosuke Yamagishi, Shouhei Hanaoka

TL;DR

The proposed method combines state-of-the-art image classification techniques, asymmetric loss for handling class imbalance, and view-based prediction aggregation to enhance classification performance.

Abstract

In this work, we present our solution for the MICCAI 2024 CXR-LT challenge, achieving 4th place in Subtask 2 and 5th in Subtask 1. We leveraged an ensemble of ConvNeXt V2 and MaxViT models, pretrained on an external chest X-ray dataset, to address the long-tailed distribution of chest findings. The proposed method combines state-of-the-art image classification techniques, asymmetric loss for handling class imbalance, and view-based prediction aggregation to enhance classification performance. Through experiments, we demonstrate the advantages of our approach in improving both detection accuracy and the handling of the long-tailed distribution in CXR findings. The code is available at https://github.com/yamagishi0824/cxrlt24-multiview-pp.

Ensemble of ConvNeXt V2 and MaxViT for Long-Tailed CXR Classification with View-Based Aggregation

TL;DR

The proposed method combines state-of-the-art image classification techniques, asymmetric loss for handling class imbalance, and view-based prediction aggregation to enhance classification performance.

Abstract

In this work, we present our solution for the MICCAI 2024 CXR-LT challenge, achieving 4th place in Subtask 2 and 5th in Subtask 1. We leveraged an ensemble of ConvNeXt V2 and MaxViT models, pretrained on an external chest X-ray dataset, to address the long-tailed distribution of chest findings. The proposed method combines state-of-the-art image classification techniques, asymmetric loss for handling class imbalance, and view-based prediction aggregation to enhance classification performance. Through experiments, we demonstrate the advantages of our approach in improving both detection accuracy and the handling of the long-tailed distribution in CXR findings. The code is available at https://github.com/yamagishi0824/cxrlt24-multiview-pp.

Paper Structure

This paper contains 24 sections, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset and Tasks
  4. Task 1: Long-tailed Classification on Large, Noisy Test Set
  5. Task 2: Long-tailed Classification on Small, Manually Annotated Test Set
  6. Model Architecture
  7. Domain-Specific Pretraining
  8. Addressing Class Imbalance
  9. Implementation Details
  10. Multi-View Aggregation
  11. Aggregation Strategy
  12. Data Preprocessing and Augmentation
  13. Training Process
  14. Training Details
  15. Evaluation Metric
  16. ...and 9 more sections

Figures (1)

  • Figure 1: Multi-View Aggregation Process for Chest X-ray Prediction. The diagram illustrates the method of combining predictions from multiple chest X-ray views to produce a final prediction value. Probabilities from frontal view(s) and lateral view(s) are processed through view-specific averaging, followed by weighted averaging across different views to generate the final prediction score.