Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Lung Ultrasound Severity Scoring Using Dedicated Feature Extractor

Jiaqi Guo, Yunan Wu, Evangelos Kaimakamis, Georgios Petmezas, Vasileios E. Papageorgiou, Nicos Maglaveras, Aggelos K. Katsaggelos

TL;DR

This work tackles the challenge of robust lung ultrasound severity scoring with limited annotated data by introducing MeDiVLAD, a pipeline that pretrains a Vision Transformer via self-supervised learning (DINO) on unlabeled frames and fuses frame features through dual-level VLAD aggregation for video-level scoring. The method uses an optional task-specific finetuning step on a small labeled set, yielding strong frame- and video-level performance while providing interpretable attention maps that highlight relevant LUS features such as A-lines, B-lines, and consolidations. Experimental results show that self-distillation can surpass fully supervised baselines at the frame level, and that MeDiVLAD outperforms common aggregation methods at the video level, approaching or exceeding the performance of finetuned CNN baselines with far fewer labels. Overall, MeDiVLAD offers a data-efficient, interpretable solution for real-time LUS severity scoring and has potential applicability to broader medical video classification tasks.

Abstract

With the advent of the COVID-19 pandemic, ultrasound imaging has emerged as a promising technique for COVID-19 detection, due to its non-invasive nature, affordability, and portability. In response, researchers have focused on developing AI-based scoring systems to provide real-time diagnostic support. However, the limited size and lack of proper annotation in publicly available ultrasound datasets pose significant challenges for training a robust AI model. This paper proposes MeDiVLAD, a novel pipeline to address the above issue for multi-level lung-ultrasound (LUS) severity scoring. In particular, we leverage self-knowledge distillation to pretrain a vision transformer (ViT) without label and aggregate frame-level features via dual-level VLAD aggregation. We show that with minimal finetuning, MeDiVLAD outperforms conventional fully-supervised methods in both frame- and video-level scoring, while offering classification reasoning with exceptional quality. This superior performance enables key applications such as the automatic identification of critical lung pathology areas and provides a robust solution for broader medical video classification tasks.

Efficient Lung Ultrasound Severity Scoring Using Dedicated Feature Extractor

TL;DR

This work tackles the challenge of robust lung ultrasound severity scoring with limited annotated data by introducing MeDiVLAD, a pipeline that pretrains a Vision Transformer via self-supervised learning (DINO) on unlabeled frames and fuses frame features through dual-level VLAD aggregation for video-level scoring. The method uses an optional task-specific finetuning step on a small labeled set, yielding strong frame- and video-level performance while providing interpretable attention maps that highlight relevant LUS features such as A-lines, B-lines, and consolidations. Experimental results show that self-distillation can surpass fully supervised baselines at the frame level, and that MeDiVLAD outperforms common aggregation methods at the video level, approaching or exceeding the performance of finetuned CNN baselines with far fewer labels. Overall, MeDiVLAD offers a data-efficient, interpretable solution for real-time LUS severity scoring and has potential applicability to broader medical video classification tasks.

Abstract

With the advent of the COVID-19 pandemic, ultrasound imaging has emerged as a promising technique for COVID-19 detection, due to its non-invasive nature, affordability, and portability. In response, researchers have focused on developing AI-based scoring systems to provide real-time diagnostic support. However, the limited size and lack of proper annotation in publicly available ultrasound datasets pose significant challenges for training a robust AI model. This paper proposes MeDiVLAD, a novel pipeline to address the above issue for multi-level lung-ultrasound (LUS) severity scoring. In particular, we leverage self-knowledge distillation to pretrain a vision transformer (ViT) without label and aggregate frame-level features via dual-level VLAD aggregation. We show that with minimal finetuning, MeDiVLAD outperforms conventional fully-supervised methods in both frame- and video-level scoring, while offering classification reasoning with exceptional quality. This superior performance enables key applications such as the automatic identification of critical lung pathology areas and provides a robust solution for broader medical video classification tasks.
Paper Structure (9 sections, 3 equations, 3 figures, 3 tables)

This paper contains 9 sections, 3 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data & Problem Statement
  3. Method
  4. Self-distillation with Task-specific Finetuning
  5. Dual-level VLAD Aggregation
  6. Experiments
  7. Image-level Classification
  8. Video-level Classification
  9. Conclusion & Discussion

Figures (3)

  • Figure 1: Score distribution of our LUS dataset, with different colors indicating different data sources. CoCross is our private dataset, other sources are detailed in COVIDxUS2021
  • Figure 2: Overall structure of MeDiVLAD, including (a) DINO-based pretraining on unlabeled frames, (b) optional task-specific finetuning on a small set of labeled frames, and (c) dual-level VLAD aggregation for video-level severity scoring.
  • Figure 3: Typical visualization of attention maps extracted from our pretrained ViT-S/8 (after finetuning).