ALGORITHM FOR FEATURE EXTRACTION OF CHROMOSOME DIGITAL IMAGES THROUGH SEGMENTATION

Abstract

This paper presents a novel algorithm for the extraction of significant features from digital images of chromosomal objects. The main goal of the algorithm is to facilitate effective clustering and identification of chromosomes based on their segmented image data. The proposed method relies on advanced image segmentation techniques that isolate chromosomal objects regardless of their geometric form, which often varies unpredictably due to biological and technical imaging factors.

A key advantage of this algorithm lies in its robustness against geometrical variability: it demonstrates consistent results even when applied to chromosome images of different shapes and contours. This adaptability makes the algorithm especially useful in real-world cytogenetic analysis, where image irregularities are common and can negatively impact the performance of classical neural networks or static feature extraction methods.

The effectiveness and precision of the developed algorithm have been rigorously evaluated through comparative analysis with the widely used convolutional neural network model VGG16. The results show that the proposed algorithm performs on par with, and in some cases even surpasses, VGG16 in terms of feature extraction quality and stability across variable datasets. This suggests that the method can be a valuable alternative or complementary approach in automated chromosome recognition systems, particularly where classical models may face limitations due to shape variability or insufficient training data.

The findings of this research contribute to the fields of digital cytogenetics, biomedical image processing, and intelligent diagnostic systems, highlighting a pathway toward more reliable chromosome analysis through tailored algorithmic approaches.