Practical application of ensemble methods for the analysis and classification of x-ray images

This study evaluates the feasibility of applying the Random Forest algorithm, as a representative of ensemble methods, for the classification of low-quality X-ray images. Image processing is computationally intensive, and the use of neural networks increases hardware requirements further. Therefore, investigating simplified approaches for such analyses is of practical importance. A series of experiments were conducted to examine various configurations of mathematical models. They were trained using the Random Forest algorithm on images of various quality (resolution ranged from 299&times;299 to 2&times;2 pixels). The findings revealed that the minimum threshold for accurate recognition corresponds to X-ray images with a resolution of 128&times;128 pixels. This level of image quality ensures compliance with the model efficiency criterion, measured by the <i>F</i>1 score, which meets or exceeds 0.83 as established through a review of related works. Additional conclusions from the study include evidence against the use of principal component analysis (PCA) method and image masks to enhance model training speed.