Automatic Rib Detection in Chest Radiographs

C. M. Brace, J. H. Kulick, T. W. ChallisIJCAI 1977

TL;DR

This paper presents an efficient algorithm for automatically detecting rib borders in chest X-rays. By cleverly avoiding global image preprocessing and leveraging knowledge of rib cage structure, it significantly reduces computational cost while achieving accurate rib localization, crucial for subsequent analysis of intercostal spaces.

What Problem Does This Paper Address?

Analyzing chest X-rays often involves examining the intercostal spaces between ribs. To do this effectively, an accurate identification of the rib borders is essential. Traditional methods for automatic rib detection, while functional, often involve computationally expensive global image enhancement and processing of every pixel, making them inefficient for large-scale screening systems. This paper aims to develop a more computationally efficient algorithm that can reliably detect rib borders, providing the necessary input for intercostal space analysis.

Methods

The proposed algorithm takes a highly optimized approach to rib detection. Instead of enhancing the entire image, it focuses on processing only the necessary information. The core strategy involves:

  • Targeted Search: The algorithm first identifies candidate areas for "starting ribs" in a pre-defined section of the lung field. This involves a multi-step process including hysteresis smoothing and a correlation-like function (MRIB) to identify potential border points.
  • Leveraging Structure: Once two adjacent starting ribs are identified, their positions and an average rib width are used to guide the detection of subsequent ribs. This is done by analyzing vertical slices through the lung field.
  • Polynomial Fitting: Detected border points are then linked together and fitted with second-degree polynomials, providing a concise representation of each rib border.
  • Efficiency: The algorithm avoids processing every pixel and only considers points likely to be rib borders, significantly reducing computational load.

Datasets

Not discussed in the paper. The paper mentions testing the algorithm on "a set of chest radiographs" and illustrates results with figures, but does not specify the dataset used or any new benchmarks.

Key Insights and Ideas

  • Computational Efficiency through Targeted Processing: The most significant insight is that global image enhancement and pixel-by-pixel processing are often unnecessary for rib detection. By focusing on specific regions and leveraging structural knowledge, computational cost can be drastically reduced.
  • Knowledge-Driven Search: Using prior knowledge about the rib cage (e.g., approximate spacing and width) as a guide for the detection process is key to the algorithm's efficiency and accuracy.
  • Polynomial Representation: Representing rib borders as second-degree polynomials offers a compact and useful output for downstream analysis.

Key Results

The algorithm demonstrates success in detecting rib borders, with typically 7 pairs of ribs identified in good quality films. The paper illustrates the results by superimposing the detected rib borders onto the original X-ray images, showing good qualitative agreement.

Limitations and Future Work

The authors acknowledge several limitations that can lead to detection failures:

  • Clavicle Interference: The clavicle's image can be mistaken for rib borders, especially in images with dark lung fields.
  • Diaphragm and Breast Tissue: In areas where the bottom rib is close to the diaphragm, or where breast tissue is present, detection can be compromised.
  • Heart Border Confusion: The heart's border can sometimes be detected as a rib border, leading to inaccurate curves.
  • Dark Areas and Gaps: If ribs are not clearly visible in dark lung fields or have significant gaps, border points may be missed or incorrectly linked.
  • Disease Obscuration: Severe disease that obscures ribs significantly impacts the algorithm's performance.

Future work is not explicitly detailed, but the success in efficient rib detection suggests its utility in automated screening systems for further analysis of intercostal spaces.

Summary updated

BibTeX
@inproceedings{brace1977automatic,
  author = {C. M. Brace and J. H. Kulick and T. W. Challis},
  title = {{Automatic Rib Detection in Chest Radiographs}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1977},
}
Figure