A Novel Method for Predicting and Diagnosing Fetal Growth Restriction: Texture Analysis Based on Ultrasound Images of Placenta During the Second Trimester (20-24 weeks).
Authors
Affiliations (4)
Affiliations (4)
- Department of Medical Ultrasonics, The Eighth Affiliated Hospital of Sun Yat-sen University, Shenzhen 518033, China.
- School of Computer Science and Technology, Dongguan University of Technology, Dongguan, 523808, China.
- School of Computer Science and Technology, Dongguan University of Technology, Dongguan, 523808, China. Electronic address: [email protected].
- Department of Medical Ultrasonics, The Eighth Affiliated Hospital of Sun Yat-sen University, Shenzhen 518033, China. Electronic address: [email protected].
Abstract
To determine whether texture analysis based on ultrasound images of placenta can be applied to identify fetal growth restriction (FGR) before clinical diagnosis. A total of 200 ultrasound images (100 FGR and 100 normal controls) of placenta (20-24 weeks) were retrospectively collected and randomly divided into a training set and an independent test set at a ratio of 8:2 using a computer-generated random split. To ensure model stability and optimize hyperparameters rigorously, we used five-fold cross-validation exclusively on the training set. Approximately 300 texture features were extracted from placenta using the methods of the grayscale histogram, grayscale co-occurrence matrix, grayscale run-length matrix, absolute gradient, autoregressive model and wavelet transform. Then, 10 optimal features were separately selected by 3 algorithms, including the Fisher coefficient method, the method of minimizing classification error probability and average correlation coefficients and the mutual information coefficient method. After nonlinear discriminant analysis was performed to reduce feature dimensionality, an artificial neural network classifier was conducted based on the statistically most significant texture features and clinical characteristics. Receiver operating characteristic curves were used to evaluate the performance of our methods in identifying FGR fetuses. Maternal and fetal baseline characteristics were similar for the FGR and normal groups, except for fetal abdominal circumference percentile, gestational age at birth and birth weight percentile (p < 0.05). Among the 30 optimal features, 10 features showed statistically significant differences between FGR and normal fetuses. The classification accuracy based on the statistically most significant texture features (p < 0.01) and abdominal circumference percentile can reach 86.50%, and the receiver operating characteristic curve for identifying FGR showed an area under the curve of 0.89. The combination of texture analysis of placenta and abdominal circumference measurement is a noninvasive, low-cost and convenient method for predicting FGR fetuses.