Optimizing Chromosome Analysis through VGT-MS: Leveraging Visual Geometric Transformer, VGG-16, and Vision Transformer for Enhanced Abnormality Identification

Abstract Chromosomes possess genetic information about the human body and their structures look like threads in the cell nucleus. Their analysis is crucial work and it is referred to as karyotyping, in which the identifications of the abnormalities from the chromosome are carried out. An efficient approach is needed to identify the abnormalities in the chromosome, even though several methods are developed their performance for detection is not efficient and encounters more problems like being time-consuming, feature extraction is not effective, etc. Therefore an effective model named Visual Geometric Transformer-based Mantis search (VGT-MS) algorithm is proposed to perform abnormality detection by overcoming the above mentioned issues. In the beginning, two datasets are exploited in the work, which comprise more chromosome images. The collection of images is carried out utilizing those two datasets such as the Chromosome karyotype Images dataset and CRCN-NE Chromosomes DataSet. But, these images contain various unnecessary things that should be eliminated. In this context, four pre-processing steps are included for the pre-processing such as Noise elimination, Contrast enhancement, Image resizing, and Normalization. After this, the VGG-16 model is deployed for making the successful feature extraction, and the Vision Transformer for the identification of the Chromosome abnormality is utilized. An initial strategy-based Mantis Search Algorithm is adopted for the hyperparameter tuning in this work. The evaluation of the developed model’s effectiveness is accomplished through F1-score, accuracy, recall, ROC, and precision where its performance is the finest performance among existing methods.