Feature Fusion Classifier With Dynamic Weights for Abnormality Detection of Amniotic Fluid Cell Chromosome

Chromosomal karyotype is important to determine whether a newborn has genetic disorder. There are two main categories of chromosomal abnormalities, structural abnormalities in which the chromosome structure altered, and number abnormalities. Manual karyotyping complex takes lot time because it requires high degree domain expertise. Based on this investigation proposes new method defect detection based deep learning with 20,299 images from Dongguan Kanghua Hospital as data integrates diversity features, trains propose classifier model feature fusion for abnormality detection. We put forward dynamic weights (FFCDW) detection, after augmentation three networks, ResNet, SENet, VGG19, trained models combined using weighting approach. Experiments prove FFCDW outperforms these mainstream VGG19. The proposed achieves precision 0.8902 F1-score 0.8805 small standard deviation (0.00903 0.00892, respectively). In addition, algorithm can automatically assign results single model, strategy fixed classifier.