An efficient deep learning approach to predict heart failure from image data using ejection fraction

Abstract

Heart is the core of human body. A normal heart beats almost 1,15,200 times in a day and 80 beats per second to make us live alive. But we often take it granted and do uncertain thinks which stops it to function perfectly. In today’s world cardiovascular diseases(CVDs) almost kill 17-18 million life’s each year worldwide which makes it the biggest disease of death. If early detection of heart malfunction or Heart failure(HF) can be detect millions of people will able to breath even longer than usual. In our research our main aim is to create an automated Deep Learning based model which will predict HF and the depth of the condition. Moreover, using which type of cardiac MRI image slice we can get better result will be consider to be our main research goal. For this we choose a cardiac MRI dataset which consists of 1100 different heart patients image having different slices in different pattern. Furthermore, with more observation and leveling different parameter with the help of Ejection Fraction(EF) values which depends on systole diastole value of heart we able to predict the heart failure with an efficient result. AI, ML & deep learning is the new trend for solving real life human problems. We used different Convolution Neural Network architecture and obtained accuracy are VGG-16(88.15%), VGG- 19(87.93%), ResNet-50 (75.85%), ResNet-101 (79.53%) Inception-V3 (85.27%). Our model is being used to find the suitable result to detect the Heart Failure(HF) with Ejection Fraction(EF).