Application of Dynamic Thermogram for Diagnosis of Hypertension

Abstract

Hypertension (high blood pressure) is blood pressure above than 140 over 90 mmHg (millimeters of mercury). Diagnosis of hypertension is made when one of the above readings is high. By the year 2025, the number of people living with hypertension is about 1.56 billion all over the world. This paper aims at developing a technique to diagnose hypertension noninvasively without using the cuff. In this approach, the dynamic infrared (IR) thermogram of selected body regions like hand (left and right) and neck (left and right) is obtained for about 60 s using IR thermal camera from 50 subjects (normal = 25 and age and sex-matched hypertensive = 25). The average temperature for every millisecond in these selected body regions is measured using ResearchIR software. Correlation is performed between the features extracted from dynamic thermogram and flow rate obtained from carotid Doppler ultrasound scan. The statistical analysis shows that highest correlation is obtained between the rate of temperature change (°C/min) in the neck left side with systolic pressure and hand left side with diastolic pressure (mmHg) (Pearson correlation r = −0.637 and 0.668 with p < 0.01, respectively). There also exists a linear correlation coefficient between neck right rate of change in temperature (°C/min) and right carotid artery flow rate (r = −0.358 with p < 0.05). An automated classifier using SVM network was designed with features extracted from dynamic thermogram for diagnosis of hypertension. The accuracy of the classifier was about 80% with sensitivity and specificity values 76.9 and 83.3%, respectively. The accuracy of the classifier when all the correlated features (n = 17) were included was 93% (sensitivity 90% and specificity 94%), whereas when highly correlated features (n = 15) were alone included, the sensitivity improved to 94% (accuracy 90% and specificity 85%).