A New Approach in Detectability of Microcalcifications in the Placenta during Pregnancy Using Textural Features and K-Nearest Neighbors Algorithm

• Published in: Journal of imaging 2022-03, Vol.8 (3), p.81
• Main Authors: Miron, Mihaela, Moldovanu, Simona, Ștefănescu, Bogdan Ioan, Culea, Mihai, Pavel, Sorin Marius, Culea-Florescu, Anisia Luiza
• Format: Article
• Language: English
• Subjects: Accuracy; Aging; Algorithms; calcifications; Classification; Feature extraction; Feature selection; Fetuses; first order feature; Fractal analysis; Fractal geometry; Fractals; Homogeneity; Hypertension; K-nearest neighbors algorithm; Kurtosis; Medical imaging; Noise; Placenta; Pregnancy; preterm placental calcifications; Prostate; second order feature; Statistical analysis; Student's t-test; Support vector machines; t-test; Ultrasonic imaging; Visibility
• ISSN: 2313-433X; 2313-433X
• DOI: 10.3390/jimaging8030081

(1) Background: Ultrasonography is the main method used during pregnancy to assess the fetal growth, amniotic fluid, umbilical cord and placenta. The placenta's structure suffers dynamic modifications throughout the whole pregnancy and many of these changes, in which placental microcalcifications are by far the most prominent, are related to the process of aging and maturation and have no effect on fetal wellbeing. However, when placental microcalcifications are noticed earlier during pregnancy, they could suggest a major placental dysfunction with serious consequences for the fetus and mother. For better detectability of microcalcifications, we propose a new approach based on improving the clarity of details and the analysis of the placental structure using first and second order statistics, and fractal dimension. (2) Methods: The methodology is based on four stages: (i) cropping the region of interest and preprocessing steps; (ii) feature extraction, first order-standard deviation (SD), skewness (SK) and kurtosis (KR)-and second order-contrast (C), homogeneity (H), correlation (CR), energy (E) and entropy (EN)-are computed from a gray level co-occurrence matrix (GLCM) and fractal dimension (FD); (iii) statistical analysis (t-test); (iv) classification with the K-Nearest Neighbors algorithm (K-NN algorithm) and performance comparison with results from the support vector machine algorithm (SVM algorithm). (3) Results: Experimental results obtained from real clinical data show an improvement in the detectability and visibility of placental microcalcifications.