METHOD OF PROCESSING DIGITAL IMAGES OF FINGERS FOR COLORIMETRIC ANALYSIS OF HEMOGLOBIN LEVEL IN BLOOD Russian patent published in 2024 - IPC G01N33/49 

FIELD: image processing means.

SUBSTANCE: present invention relates to means of processing data obtained during non-invasive analysis of hemoglobin content in patient's blood using machine learning models. Disclosed is a method of processing digital images of fingers for colorimetric analysis of hemoglobin level in blood, which consists in the fact that on the digital image of the fingers, the image areas on which the nail beds are located, as well as the skin areas, the color of which changes depending on the blood hemoglobin level; wherein a portion of the image of the nail bed and the skin area in which the distribution of intensity in the different color channels is uniform is segmented; filtering the segmented image regions, the color characteristics of which vary depending on the hemoglobin level in the blood; while removing areas in which intensities of R-, G-, B-channels have atypical characteristics for color characteristics of skin and nail bed by comparing average intensities and standard deviation of intensities of the detected image with reference values; for filtered image areas, statistics of distribution of intensities are calculated simultaneously in R-, G-, B-channels of a digital image: percentiles of distribution of intensities at level of 5, 15, 25, 50, 75, 85, 95 of the average value of intensity and standard deviation of intensities; hemoglobin concentration is predicted using a machine learning model based on an ensemble of decision trees based on calculated values of statistics of intensity distribution of the digital image of the selected areas.

EFFECT: present invention enables more accurate recognition of the obtained images of the nail beds of the patient's fingers by taking into account the distribution of light intensities passing through the patient's tissue, the obtained digital image simultaneously in all three R-, G-, B-channels, using at that several levels of values of specified reference values of distribution of intensities and such a machine learning algorithm, which will make it possible to more fully take into account the above distribution of intensities and more accurately predict the value of hemoglobin based on the analysis of the obtained RGB images.

3 cl, 4 dwg