1Dept. of Computer Engineering, JSPM's Rajarshi Shahu College of Engineering, S.P.P.U, Pune, India
2Professor, Dept. of Electronics and Telecommunication, AISSMS College of Engineering, Pune, India
Online published on 13 December, 2018.
Hypertension is leading cause of deaths all over the world as it leads to heart, kidney and brain damage. Most of the Indian population is not health conscious. Around 58% urban and 75% rural Indians are not aware of their hypertensive status. As per survey in 2013, there is increase of 138% in deaths due to hypertension in India in comparison with number of deaths in 1990. This paper presents an approach for predicting hypertension using handwritten manuscript which is based on science of handwriting analysis. Handwriting analysis is the ancient science which is used to predict personality, emotional state and health conditions of the writer. As per the science of handwriting analysis, the handwriting stokes of the person remains same throughout the life. Handwriting is recognized as being unique to each individual. It is not related to gender and age. Handwriting analysis is used to understand PRE-ILLNESS warnings which appear in the handwriting much before the disease symptoms can be detected by any modern equipment or tests. In this paper, an approach for predicting hypertensive people based on handwritten manuscript is presented. The proposed system extracts eighteen writing features such as Euler number, number of right and left slant lines, number of horizontal and vertical lines, total length of horizontal and vertical lines, total length of left and right slant lines, overall image size, top margin, right margin, bottom margin, left margin, spacing between the words and lines, font size, and letter slant from a handwritten manuscript. Three different feature sets namely geometric, graphological and integrated feature set have been formed using extracted writing features.
This study uses a dataset of 150 handwritten manuscripts of hypertensive people and 150 of control group people. The performance of the system has been evaluated with different size datasets and for each feature set.
The proposed approach provides maximum 93.30% sensitivity, 92.2% specificity, 92.75% accuracy, 92.25% positive predictive value, 93.30% negative predictive value, positive likelihood ratio 12.52, negative likelihood ratio 0.07 using support vector machine classifier with integrated feature set.
Extremely useful in healthcare centers and individual health analysis.
Handwritten manuscript, writing features, handwriting analysis, geometric features, graphological features, integrated feature set
