Pretrained DcAlexnet Cardiac Diseases Classification on Cognitive Multi-Lead Ultrasound Dataset

Abstract

The DcAlexNetCNN deep learning classifier can easily track patterns in medical images (brain, heart, spinal cord and etc.) precisely. According to WHO (world health organization) every year 5 billion people are affecting heart diseases and heart-attacks. Heart abnormalities sometimes tends to death; therefore, an efficient medical image pre-processor and deep learning classifier is needed for diagnosis. So that in this research work multi-class DcAlexNet classifier, RRS-HSB segment-filter has been implemented. The RRS (Restrictive Random segmentation) and GHSB (Gaussian Hue saturation brightness filtration) modules are fused to get multi-level feature. The training process has been incorporated to EchoNet dataset and testing process can be verified to real time samples. The segmented features as well as filtered feature are loaded into weighted .CSV file. The following features are classified tends to get predicting abnormalities in heart ultra sound image. The pretrained DcAlexNet CNN model is loading to EchoNet 1 lakh samples using 165 layers such as normalized layer, dense layer, flatten layer, max pooling layer and ReLu layer. The computer aided design with corresponding CNN layers has been finding hidden sample over to get heart abnormality location. The experimental results in terms of Dice score 98.89%, Accuracy 99.455, precision 99.23%, recall 98.34%, F-1 score 98.92%, CC 99.27%, and sensitivity 99.34% had been attained. The attained performance metrics are competed with present technologies and outperformance the application accuracy on heart diagnosis.