Image Processing to Quantitate Hemoglobin Level for Diagnostic Support
Main Article Content
Abstract
Digital Image processing techniques can be used to explore primitive diagnostics methods for
disease detection at early stages with limited resources and skilled labor. These techniques can
also assist doctors during clinical examination without any need for invasive pathological test and
this facilitates patient comfort and avoids infection during blood test. Various blood components
such as hemoglobin and billirubin whose approximate measure can directly be identified by just
viewing the color of patient skin, nails, eye or any other target area can be measured and classified
in terms of the color content of the image of the targeted area. Analysis of image processing
techniques in conjunction with specialized supervision can provide significant exploration in the
field of biomedicine and clinical applications. This research work proposes an image processing
based non-invasive method of measuring hemoglobin (Hb) concentration present in patient’s
blood by analyzing the color and texture of digital photographs of patient’s palpebral conjunctiva.
The images of patient’s palpebral conjunctiva were processed and eight relevant features were
extracted .Artificial neural network classifier was used to correlate the output quantity to be
measured with the values of the quantity measured by the standard method as per the guidelines
given by WHO. Further, based on the testing results obtained by the classifier the patients whose
Hb concentration was less than 11g/dL were screened as anaemic patients. A confusion matrix
was then plotted to evaluate and compare the predicted classification results with the actual value
of Hb obtained from invasive test. It was found that the proposed algorithm was able to diagnose
anemia with 71.42% sensitivity and 89.47% specificity. The proposed method is helpful for
detection of not only severe anemia but works well in detection of moderate anemia too thus
predicting the hemoglobin value to an accuracy of 81.81%.The proposed work is useful for giving
assistance to medical practitioners for reliable diagnosis of anaemia in the clinic itself and in low
resource settings.