Abstract
This work proposes the use of Genetic Algorithms (GA) in tracing and recognizing the pericardium contour of the human heart using Computed Tomography (CT) images. We assume that each slice of the pericardium can be modelled by an ellipse, the parameters of which need to be optimally determined. An optimal ellipse would be one that closely follows the pericardium contour and, consequently, separates appropriately the epicardial and mediastinal fats of the human heart. Tracing and automatically identifying the pericardium contour aids in medical diagnosis. Usually, this process is done manually or not done at all due to the effort required. Besides, detecting the pericardium may improve previously proposed automated methodologies that separate the two types of fat associated to the human heart. Quantification of these fats provides important health risk marker information, as they are associated with the development of certain cardiovascular pathologies. Finally, we conclude that GA offers satisfiable solutions in a feasible amount of processing time.
Introduction
Quantifying the fat surrounding the human heart is a vital health marker. Clinicians specifically distinguish between epicardial fat (located inside the pericardium) and mediastinal fat (located outside). High volumes of these adipose tissues are strongly correlated with the development of atherosclerosis, coronary artery disease, and other significant cardiovascular risks.
The primary challenge in automating this quantification is the "pericardium contour"—the thin membrane separating these two fat types. In CT images, this boundary is notoriously difficult to distinguish automatically. Traditionally, radiologists have traced it manually, a process that is tedious and prone to inter-observer variability. Previous automated attempts using complex machine learning models were accurate but computationally expensive, often taking nearly two hours to process a single patient’s data. This study proposes a Genetic Algorithm (GA) approach to find this boundary quickly and accurately by modeling it as a geometric primitive.
Methodology
The core of the proposed method is the use of a Genetic Algorithm to fit an ellipse to the pericardium. In the axial plane of a CT scan, the pericardium naturally approximates an elliptical shape. By defining the problem as a search for optimal ellipse parameters, the GA can efficiently navigate the search space.
The GA evolves a population of candidate ellipses, where each individual is defined by five parameters:
- Rotation angle ($\theta$)
- Center coordinates ($x_c, y_c$)
- Major axis length ($a$)
- Minor axis length ($b$)
The Fitness Function
The fitness function is the critical component that determines how well a candidate ellipse fits the actual pericardium. The algorithm evaluates the pixels enclosed by the ellipse using a weighted scoring system based on pre-processed "fat-ranged" images:
- Positive weight: Red pixels, representing epicardial fat. The goal is to maximize the inclusion of these pixels.
- Negative weights: Green pixels (mediastinal fat), grey pixels (other tissues), and black pixels (background). The goal is to exclude these from the ellipse.
Standard GA operators—selection, crossover, and mutation—are used to evolve the population over a set number of generations (typically 10, 100, or 200) until an optimal ellipse that "hugs" the pericardium boundary is identified.
Datasets and Pre-processing
The study utilized a dataset of 20 patients, totaling 878 CT images. To provide a "gold standard" for comparison, the images were manually segmented by experts. Before applying the GA, images were filtered to a specific Hounsfield scale range (-200 to -30 HU) to isolate fat tissues, resulting in the color-coded images used by the fitness function.
Key Results and Insights
The researchers evaluated the algorithm across different generation counts ($n$) to balance accuracy and computational speed:
- Accuracy: At 200 generations, the algorithm captured an average of 99.45% of the target epicardial fat (Red pixels) inside the ellipse.
- General Fit (GF): The GF index, which measures the ratio of desired fat to undesired pixels, improved from 1.95 at 10 generations to 3.37 at 200 generations.
- Efficiency: While previous machine learning approaches required approximately 1.8 hours per patient, the GA approach processes a single image in roughly 13 minutes. By propagating the optimal ellipse from one slice to the remaining ~43 images of a patient, the total processing time per patient is estimated to drop to approximately 0.6 hours.
Conclusion
This study demonstrates that using geometric primitives like ellipses as shortcuts for complex boundary detection is a viable and efficient clinical strategy. The GA allows for a robust search of a massive parameter space far faster than brute-force methods. While the elliptical assumption is a simplification—as the pericardium is not perfectly elliptical in every slice—the results are satisfiable for clinical fat quantification. Future work will focus on refining the algorithm to handle non-elliptical irregularities and utilizing regression algorithms to estimate total fat volume from a reduced number of slices.
Summary updated