Obstructive Pulmonary Disease Prediction Through Heart Structure Analysis

Abstract

There are more than 200 million cases of lung diseases worldwide. Most commonly they include Obstructive Pulmonary Disease (OPD) like Chronic OPD (COPD), asthma and bronchiectasis. This paper aims to study the causal relationship between the shape of the heart and presence of obstructive pulmonary disease by analyzing HRCT scans which are sensitive and informative as it provides with multiple slices of a patient’s internal structure. A mathematical model to predict disease gives confidence to the radiologists for correct and early diagnosis of the disease. Real life HRCT scans along with the disease information were obtained from the Institute of Pulmocare and Research (IPCR). Using Image Processing techniques we finally obtained the right and left atrium of the heart as individual gray-scale images from the HRCT scans; which were then converted into a gray-scale matrix and finally into a vector. We generated our data-set consisting of 40 patients. For patient diagnosed with Obstructive Pulmonary Disease we assigned the label as +1 and for those who have other disease as −1. Different machine learning algorithms such as kNN, SVM, Random Forest and Naive Bayes were applied to the dataset to find the algorithm with highest accuracy and maximum area under the ROC plot.