DEEP LEARNING PLACENTA ABNORMALITIES WITH CUSTOMIZED ARCHITECTURE USING DEEPLEARNING4J

Abstract

The conditions at the origin place of physical birth namely placenta is normally analyzed and determined by physicians as well as by the support of Image Processing techniques during pregnancy. Such support is measured by the performance metrics like Accuracy, ROC, Precision, Recall, and F-Measure generated by the models based on Machine Learning. In this study we apply the customized approach implemented in Weka tool, especially the package DeepLearning4j..They are found to be comparable with the usage of standard architectures like LeNet, VGGnet, ResNet, and Alexnet. Quality of the proposed architecture is tested and refined for optimization using DeepLearning4j, iterated over both the Loss functions set at input layer as well as the errors measured at the output layer. It has been found the experimental results using ‘MCXENT’ function, denoting ‘Multi-class Cross Entropy’ among selected seven different loss functions generates more accuracy and less error. And the best performance is achieved with the maximum accuracy of 95.7%. These results support and present a new confidence for yet another machine learning approach using interactive development tool for gynecologists.