Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Volumes to Slices: Computationally Efficient Contrastive Learning for Sequential Abdominal CT Analysis

Po-Kai Chiu, Hung-Hsuan Chen

TL;DR

This work tackles label scarcity in abdominal CT analysis by introducing 2D-VoCo, a computationally efficient slice-level self-supervised learning framework inspired by 3D VoCo. The method uses a momentum-based student-teacher setup with an EfficientNetV2 backbone to learn spatial-semantic features from 2D CT slices via intra-, inter-, and regularization losses, then transfers the learned representation to a CNN-LSTM classifier for multi-organ injury assessment. Experimental results on the RSNA 2023 Abdominal Traumatic Injury dataset show that 2D-VoCo pre-training improves RSNA score, mAP, precision, and recall over ImageNet pre-training, with further gains when incorporating large unlabeled datasets like FLARE23 and when usefully modeling full abdominal scans (multi-organ) rather than single-organ ROIs. The findings support its practicality for reducing labeled-data dependence and enhancing clinical CT analysis, and code is released for reproducibility. The work highlights the value of slice-level SSL for complex anatomical tasks and opens avenues for richer sequence modeling and interpretability in clinical pipelines.

Abstract

The requirement for expert annotations limits the effectiveness of deep learning for medical image analysis. Although 3D self-supervised methods like volume contrast learning (VoCo) are powerful and partially address the labeling scarcity issue, their high computational cost and memory consumption are barriers. We propose 2D-VoCo, an efficient adaptation of the VoCo framework for slice-level self-supervised pre-training that learns spatial-semantic features from unlabeled 2D CT slices via contrastive learning. The pre-trained CNN backbone is then integrated into a CNN-LSTM architecture to classify multi-organ injuries. In the RSNA 2023 Abdominal Trauma dataset, 2D-VoCo pre-training significantly improves mAP, precision, recall, and RSNA score over training from scratch. Our framework provides a practical method to reduce the dependency on labeled data and enhance model performance in clinical CT analysis. We release the code for reproducibility. https://github.com/tkz05/2D-VoCo-CT-Classifier

From Volumes to Slices: Computationally Efficient Contrastive Learning for Sequential Abdominal CT Analysis

TL;DR

This work tackles label scarcity in abdominal CT analysis by introducing 2D-VoCo, a computationally efficient slice-level self-supervised learning framework inspired by 3D VoCo. The method uses a momentum-based student-teacher setup with an EfficientNetV2 backbone to learn spatial-semantic features from 2D CT slices via intra-, inter-, and regularization losses, then transfers the learned representation to a CNN-LSTM classifier for multi-organ injury assessment. Experimental results on the RSNA 2023 Abdominal Traumatic Injury dataset show that 2D-VoCo pre-training improves RSNA score, mAP, precision, and recall over ImageNet pre-training, with further gains when incorporating large unlabeled datasets like FLARE23 and when usefully modeling full abdominal scans (multi-organ) rather than single-organ ROIs. The findings support its practicality for reducing labeled-data dependence and enhancing clinical CT analysis, and code is released for reproducibility. The work highlights the value of slice-level SSL for complex anatomical tasks and opens avenues for richer sequence modeling and interpretability in clinical pipelines.

Abstract

The requirement for expert annotations limits the effectiveness of deep learning for medical image analysis. Although 3D self-supervised methods like volume contrast learning (VoCo) are powerful and partially address the labeling scarcity issue, their high computational cost and memory consumption are barriers. We propose 2D-VoCo, an efficient adaptation of the VoCo framework for slice-level self-supervised pre-training that learns spatial-semantic features from unlabeled 2D CT slices via contrastive learning. The pre-trained CNN backbone is then integrated into a CNN-LSTM architecture to classify multi-organ injuries. In the RSNA 2023 Abdominal Trauma dataset, 2D-VoCo pre-training significantly improves mAP, precision, recall, and RSNA score over training from scratch. Our framework provides a practical method to reduce the dependency on labeled data and enhance model performance in clinical CT analysis. We release the code for reproducibility. https://github.com/tkz05/2D-VoCo-CT-Classifier
Paper Structure (12 sections, 3 equations, 2 figures, 2 tables)

This paper contains 12 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Upstream Pre-training: The 2D-VoCo Framework
  4. Downstream Fine-tuning: Classification
  5. Experiments
  6. Datasets and Pre-processing
  7. Experimental Setup
  8. Results and Analysis
  9. Effectiveness of 2D-VoCo Pre-training
  10. Impact of External Unlabeled Data
  11. Multi-Organ vs. Single-Organ Classification
  12. Discussion

Figures (2)

  • Figure 1: Overall Framework: the process includes an upstream pre-training stage with the 2D-VoCo student-teacher framework and a downstream fine-tuning stage using a CNN-LSTM architecture for classification.
  • Figure 2: Intra loss and regularization loss for 2D VoCo pre-training.