Detection of Fetal Hypertropic Cardiomyopathy Using Machine Learning Techniques

This problem statement was chosen as my final semester major project. This is one of that heart-related diseases whose first symptoms are death. The disease is caused by the abrupt thickening of the heart muscles. There is no cure for this disease as of yet as it's difficult to identify this disease. As a result, I decided to take up this project and I concentrated mainly on the identification of this disease in infants. HCM is a hereditary disease, so in my study, I concentrated mainly on the diabetic mothers with HCM. The insulin given to these mothers will also result in the thickening of the heart muscles of their infants. 

The block diagram of the approach is given below:

Methodology

We decided to employ 2d Multiscale entropy in this project. We chose this particular methodology because of its high precision on 1d data. We developed the code for2d data based on the original 1d code. CLAHE was also used to enhance the images. Multiscale entropy features were extracted from the Littlewood Paley coefficients. Finally, the KNN classifiers were applied and arrived at the classification accuracy of 73.8% using Cosine KNN. 

Softwares

Matlab

Challenges

This was intellectually challenging because of the freshness which demanded a thorough understanding of the database as no work had been done on this particular subject.

After my final presentation of this project, we decided to enhance the model to get higher classification accuracy, so instead of using just multiscale entropy, we combined Gabor Wavelets and Entropy features along with SVM to obtain an accuracy of 80.39%. Details of the same are available at https://tejaswi-n-rao.webnode.in/publications/ 

The block diagram of the enhanced model is shown below