Dataset Overview

Dataset Information

• Dataset Name: ChestX-Det10
• Data Source: This dataset is a subset of the NIH ChestX-14 dataset.
• Data Type: Chest X‑ray images
• Number of Images:
  • Training images: 3,001
  • Test images: 1,000+
• Number of Classes: 10
• Class List:
  • Consolidation
  • Pneumothorax
  • Emphysema
  • Calcification
  • Nodule
  • Mass
  • Fracture
  • Effusion
  • Atelectasis
  • Fibrosis
• Data Missing: 22.69% of the data is considered background.

Dataset Characteristics

• Class Distribution: The dataset exhibits class imbalance, with Consolidation and Effusion being the most common, while Mass and Pneumothorax are relatively rare.
• Bounding Box Distribution: Most bounding boxes are small, with width and height primarily below 200 pixels, and the area distribution skewed toward smaller regions.

Data Processing

• Data Augmentation: Applied random horizontal flips, random rotations, color jitter, and normalization.
• Class Weight Adjustment: Used weighted loss functions to address class imbalance.
• Attention Mechanism: Introduced CBAM (Convolutional Block Attention Module) to enhance feature selection.

Dataset Applications

• Object Detection: Used to train and evaluate YOLOv8 and Faster R-CNN models for detecting abnormalities in chest X‑ray images.
• Model Comparison: Analyze the performance of YOLOv8 and Faster R-CNN in terms of speed, accuracy, and precision.
• Performance Optimization: Optimize model performance using class weighting, data augmentation, and learning rate adjustments.

Dataset Citation

• ChestX-ray14 Dataset: arXiv:1705.02315
• CheXNet: arXiv:1711.05225