Highly Performing Automatic Detection of Structural Chromosomal Abnormalities Using Siamese Architecture

• DOI: 10.2139/ssrn.4246604
• Publication Date: 2022-10-14T17:20:25Z
• Authors: Mohammed El Amine...; Jean-Marie Guyader; Marwa El Bouz; Nathalie Douet-Gu...; Ayman Al Falou; Marie-Bérengère T...
• Keywords: Chromosome Aberrations; deletion/inversion detection; Genetic Diseases, Inborn; [INFO.INFO-CV]Computer Science [cs]/Computer Vision and Pattern Recognition [cs.CV]; 610; Datasets as Topic; cytogenetics; Chromosomes; 004; 3. Good health; siamese architecture; 03 medical and health sciences; 0302 clinical medicine; structural chromosomal abnormalities; Neoplasms; convolutional neural networks; Humans; Neural Networks, Computer

The detection of chromosome abnormalities is crucial for the diagnosis, prognosis and management many genetic diseases cancers. This detection, done by highly qualified medical experts, tedious time-consuming. We propose a performing, intelligent automatic method to assist cytogeneticists screen structural chromosomal (SCA). Each present in two copies that make up pair chromosomes. Usually, SCA are only one copy pair. Convolutional neural networks (CNN) with Siamese architecture particularly relevant evaluating similarities between images, which why we intend use this detect within pairs As proof-of-concept, first focused on deletion occurring 5 (del(5q)) observed hematological malignancies. From our dataset, highest performance detecting was provided MobileNet, achieving 98.71% accuracy 97.14% F1-score. additionally verified model, trained del(5q) normal chromosomes, able identify most difficult inversion inv(16), 66.12% accuracy, F1-score 57.87%. It performing based allows SCA.