An intelligent validation system for diagnostic and prognosis of ultrasound fetal growth analysis using Neuro-Fuzzy based on genetic algorithm

• Published in: Egyptian informatics journal Vol. 20; no. 1; pp. 55 - 87
• Main Authors: Kaur, Prabhpreet, Singh, Gurvinder, Kaur, Parminder
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2019; Elsevier
• Subjects: Artificial Neural Network (ANN); Classification learner; Computer-Aided Diagnostic (CAD); Normal-Shrink; Ultrasound (US); Wavelet; Computer-Aided Diagnostic (CAD); Wavelet; Ultrasound (US); Normal-Shrink; Artificial Neural Network (ANN); Classification learner
• ISSN: 1110-8665; 2090-4754; 2090-4754
• DOI: 10.1016/j.eij.2018.10.002

[Display omitted] Acquisition of the standard plane is the prerequisite of biometric measurement and diagnosis during the ultrasound (US) examination. Based upon the analysis of existing algorithms for the automatic fetal development measurement, a new algorithm known as Neuro-Fuzzy based on genetic algorithm is developed. Firstly, the fetal ultrasound benchmark image is auto-pre-processed using Normal Shrink Homomorphic technique. Secondly, the features are extracted using Gray Level Co-occurrence Matrix (GLCM), Grey Level Run Length Matrix (GLRLM), Intensity Histogram (IH) and Rotation Invariant Moments (IM). Thirdly, Neuro-Fuzzy using Genetic approach is used to distinguish among the fetus growth as abnormal or normal. Experimental results using benchmark and live dataset demonstrate that the developed method achieves an accuracy of 97% as compared to the state-of- art methods in terms of parameters such as Sensitivity, Specificity, Recall, F-Measure &Precision Rate. The use of area under the receiver of characteristics(AUC) and confusion matrix as assessment indicators is also cross-validated using various classification methods by achieving best accuracy rate of Support Vector Machine (SVM) i.e. 98.7% as compare to other classification methods such as KNN, Ensemble methods, Linear Discriminant Analysis(LDA) and Decision Tree whereas ROC curve covers 0.9992 SVM.