Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lung-DETR: Deformable Detection Transformer for Sparse Lung Nodule Anomaly Detection

Hooman Ramezani, Dionne Aleman, Daniel Létourneau

TL;DR

This work tackles robust detection of sparse lung nodules in CT scans by reframing the task as anomaly detection and leveraging a preprocessing pipeline with Maximum Intensity Projection (MIP). It introduces Lung-DETR, a Deformable-DET R-based framework with a focal loss variant, designed to handle extreme class imbalance and small object detection. On the LUNA16 benchmark, it achieves a high F1 score of 94.2% and strong AP/AR across size categories, especially for medium and large nodules, while maintaining low false positives. The approach demonstrates potential for real-world clinical deployment, including in resource-limited settings, by improving detection sensitivity under realistic sparsity and anatomical complexity.

Abstract

Accurate lung nodule detection for computed tomography (CT) scan imagery is challenging in real-world settings due to the sparse occurrence of nodules and similarity to other anatomical structures. In a typical positive case, nodules may appear in as few as 3% of CT slices, complicating detection. To address this, we reframe the problem as an anomaly detection task, targeting rare nodule occurrences in a predominantly normal dataset. We introduce a novel solution leveraging custom data preprocessing and Deformable Detection Transformer (Deformable- DETR). A 7.5mm Maximum Intensity Projection (MIP) is utilized to combine adjacent lung slices into single images, reducing the slice count and decreasing nodule sparsity. This enhances spatial context, allowing for better differentiation between nodules and other structures such as complex vascular structures and bronchioles. Deformable-DETR is employed to detect nodules, with a custom focal loss function to better handle the imbalanced dataset. Our model achieves state-of-the-art performance on the LUNA16 dataset with an F1 score of 94.2% (95.2% recall, 93.3% precision) on a dataset sparsely populated with lung nodules that is reflective of real-world clinical data.

Lung-DETR: Deformable Detection Transformer for Sparse Lung Nodule Anomaly Detection

TL;DR

This work tackles robust detection of sparse lung nodules in CT scans by reframing the task as anomaly detection and leveraging a preprocessing pipeline with Maximum Intensity Projection (MIP). It introduces Lung-DETR, a Deformable-DET R-based framework with a focal loss variant, designed to handle extreme class imbalance and small object detection. On the LUNA16 benchmark, it achieves a high F1 score of 94.2% and strong AP/AR across size categories, especially for medium and large nodules, while maintaining low false positives. The approach demonstrates potential for real-world clinical deployment, including in resource-limited settings, by improving detection sensitivity under realistic sparsity and anatomical complexity.

Abstract

Accurate lung nodule detection for computed tomography (CT) scan imagery is challenging in real-world settings due to the sparse occurrence of nodules and similarity to other anatomical structures. In a typical positive case, nodules may appear in as few as 3% of CT slices, complicating detection. To address this, we reframe the problem as an anomaly detection task, targeting rare nodule occurrences in a predominantly normal dataset. We introduce a novel solution leveraging custom data preprocessing and Deformable Detection Transformer (Deformable- DETR). A 7.5mm Maximum Intensity Projection (MIP) is utilized to combine adjacent lung slices into single images, reducing the slice count and decreasing nodule sparsity. This enhances spatial context, allowing for better differentiation between nodules and other structures such as complex vascular structures and bronchioles. Deformable-DETR is employed to detect nodules, with a custom focal loss function to better handle the imbalanced dataset. Our model achieves state-of-the-art performance on the LUNA16 dataset with an F1 score of 94.2% (95.2% recall, 93.3% precision) on a dataset sparsely populated with lung nodules that is reflective of real-world clinical data.
Paper Structure (13 sections, 3 equations, 3 figures, 1 table)

This paper contains 13 sections, 3 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction and Related Work
  2. Computed Tomography
  3. Handling Imbalanced Data in Deep Learning
  4. The LUNA16 Dataset
  5. Maximum Intensity Projection
  6. Detection Transformer (DETR)
  7. Method Overview
  8. Methodology
  9. Data Preprocessing
  10. Dataset
  11. Deformable-Detection Transformer
  12. Focal Loss for Classification
  13. Results

Figures (3)

  • Figure 1: Data Processing Pipeline With Tumor Visible Top Left of Lung
  • Figure 2: Lung-DETR Architecture
  • Figure 3: Lung-DETR Predicitions on Slices with Tumor