Nowadays, smartphones are used as self-health monitoring devices for humans. Self-health monitoring devices help clinicians with big data for accurate diagnosis and guidance for treatment through repetitive measurement. Repetitive measurement of haemoglobin requires for pregnant women, pediatric, pulmonary hypertension and obstetric patients. Noninvasive haemoglobin measurement through conjunctiva leads to inaccurate measurement. The inaccuracy is due to a decrease in the density of goblet cells and acinar units in Meibomian glands in the human eye as age increases. Furthermore, conjunctivitis is a disease in the eye due to inflammation or infection at the conjunctiva. Conjunctivitis is in the form of lines in the eyelid and covers the white part of the eyeball. Moreover, small blood vessels in eye regions of conjunctiva inflammations are not visible to the human eye or standard camera. This paper proposes smartphone-based haemoglobin (SBH) measurement through a borescope camera from anterior ciliary arteries of the eye for the above problem. The proposed SBH method acquires images from the anterior ciliary arteries region of the eye through a smartphone attached with a high megapixel borescope camera. The anterior ciliary arteries are projected through transverse dyadic wavelet transform (TDyWT) and applied with delta segmentation to obtain blood cells from the ciliary arteries of the eye. Furthermore, the Gaussian regression algorithm measures haemoglobin (Hb) with more accuracy based on the person, eye arteries, red pixel statistical parameters obtained from the left and right eye, age, and weight. Furthermore, the experimental result of the proposed SBH method has an accuracy of 96% in haemoglobin measurement.
Fifty percent of women of reproductive age are affected by anaemia, and continuous monitoring of haemoglobin (Hb) becomes a global priority [
Noninvasive haemoglobin measurements from various regions such as finger [
In this paper, the above problems are overcome by measuring haemoglobin from anterior ciliary arteries of the eye, above the Bulbar and palpebral conjunctiva regions. In the human body, measurement of various parameters such as heart rate and oxygen saturation (Spo2) through various sensors such as optical Light Emitting Diode (LED) and Light Amplification by Stimulated Emission of Radiation (LASER) obtains the values from blood flow in arteries. The radiation from the sensor such as LED [ The proposed SBH method increases accuracy in measuring haemoglobin due to direct measurement from arteries of the eye because arteries (in red) are the blood vessels that deliver blood to the body. The proposed SBH method uses a borescope camera to magnify the artery region in the eye and avoids errors related to an artefact such as eyeball movement such as nystagmus, vergence, saccade, pursuit, miniature during image acquisition. The proposed SBH method improves the accuracy due to the perspective projection of arteries in the eye, which varies the size of the artery inversely proportional with distance and provides the accurate image of eye arteries through TDyWT transform and Delta-E segmentation for identifying anterior ciliary artery colour.
In this paper, a borescope camera is used to acquire arteries image from the eyeball region for haemoglobin measurement because the camera has a higher degree of field of view, such as 120 degrees and brighter light, compared with an endoscope camera. Brighter light in borescope camera avoids ambient noises in eye image during acquisition. Brighter light in the borescope camera increases the distance between eye and camera during image acquisition. A higher field of view leads to a maximum area of regions in an image. This paper uses a borescope camera Model No AN98B (High Definition: 1280 x 720) for eye artery image acquisition. Borescope camera consists of six white LED lights, 5 cm to infinite of the focal distance with LED and photo control switch.
In the human eye, anterior ciliary arteries consist of 7 small arteries in each eye socket and supply blood to eye regions such as the conjunctiva, sclera and rectus muscles. Ciliary arteries are the branches of the ophthalmic artery. In the human eye, the anterior ciliary arteries are in the front side of the eyeball and extraocular muscles. They form a vascular zone beneath the conjunctiva and then pierce the sclera a short distance from the cornea and end in the circulus arteriosus major. Three of the four rectus muscles, the superior, inferior, and medial, are supplied by two ciliary arteries, while the lateral rectus only receives one branch. Ciliary arteries over the sclera region consist of noise during image acquisition due to reflections, luminosity, contrast [
A hybrid median filter reduces contrast and improves edges and features compared to a median filter. Furthermore, the hybrid median filter maintains a difference in brightness in contrast to the median filter. The non-linear hybrid mean filter does not smooth the image excessively and reduces less noise in the image. The hybrid median filter of the anterior ciliary artery image is shown in
Discrete Wavelet Transform (DWT) remove noises due to luminosity, contrast and reflections in anterior ciliary artery image [
To improve qualitative results in anterior ciliary artery image, non-decimated Wavelet transforms such as Stationary Wavelet is applied. Stationary wavelet transform (SWT) is an undecimated wavelet transform never performs coefficient decimation at each level of transformation. SWT has a redundancy property, where wavelet coefficients possess equal samples similar to the input image. The 2N redundancy requires N level decomposition; N represents decomposition level. Furthermore, SWT avoids downsampling during decomposition. Thresholding techniques in stationary wavelet transform perform on a single wavelet coefficient at a particular time. Each wavelet coefficient lesser than the threshold value, can be set as zero or further changed.
The noise predominates wavelet coefficients with a value less than a threshold value; wavelet coefficients exceeding the threshold value have the image information than noise. The advantage of wavelet thresholding includes the automatic selection of threshold level for denoising process without a prior idea of image. In the proposed SBH method, hard thresholding is performed, which involves multiplication of wavelet coefficient value larger than threshold value with standard deviation value of noise. The hard thresholding method reduces the noise level to a great extent by thresholding wavelet coefficients. SWT possess a time-invariant property that suits image denoising of anterior ciliary artery image as shown in
Filtered images of the anterior ciliary artery are applied to Dyadic Wavelet transform (DyWT) for perspective projection of arteries in the eye region. DyWT is similar to DWT; discrete wavelet transform does not possess shift-invariant property. DWT uses the downsampling method. Furthermore, during the convolution process of decomposition stage, the coefficients of the second Wavelet alone are considered for downsampling and hence the discrete Wavelet transform is said as decimated Wavelet transform. Due to the lack of shift-invariance property, DWT results in poor performance in enhancement and edge detection in the anterior ciliary artery region of the eye. Hence, for better results, dyadic wavelet transform is proposed in SBH method for perspective enhancement of arteries of the eye image. Dyadic wavelet transform, an undecimated discrete wavelet transform, comes under the redundant waveform category. Dyadic wavelet transform’s shift-invariant property improves sampling in the time domain. Moreover, the dyadic wavelet transform performs comparatively better than the DWT wavelet transform.DWT possess less compression and denoising performance. Dyadic wavelet transform is relatively flexible and attains symmetric and smoothened anterior ciliary artery region pixels. Furthermore, downsampling is avoided, dyadic Wavelet transforms suit for texture analysis. For the computation of dyadic wavelet transform of anterior ciliary artery region image, consider the decomposable image to be I and let p[k] and q[k] are the scaling and wavelet filters. Initially, for scale r = 0 and considering I0 = I, the Wavelet and scaling coefficients are computed as follows: For scales, r = 1, 2 … R, the Wavelet and scaling coefficients are given as:
Consider p
Haar decomposition effectively enhances transient features and enhances the features in the entire spectrum without a dominant frequency band. Burt reconstruction has scale-frequency and translation time shift properties for a short and big window for low scales-high frequencies and high scales-low frequencies. In traditional, decomposition and reconstruction are performed with a single wavelet family. In proposed SBH method, transverse dyadic wavelet transform performs through individual pixels or the permutation of their grey levels. In transverse dyadic shifting performs through dyadic correlations and enhances the anterior ciliary artery region image without loss of information. In transverse dyads, the local extrema play a vital role in perspective projection and enhance the edges of the anterior ciliary artery region. Local extrema are the magnitude value and correspond to the edges of the anterior ciliary artery region in the image object for enhancement. In TDyWT, lower minor edges are enhanced at lower transform levels. When the input anterior ciliary artery region image is of N x M, decomposition of Wavelet is processed for Log2 (N) levels. A desired number of iterations is performed for perspective projections. TDyWT resultant anterior ciliary artery region image provides high resolution in time for high-frequency components and more excellent resolution in frequency for low-frequency components.
Delta Segmentation shows the colour difference between a selected anterior ciliary artery and sclera or conjunctiva of the eye region. Delta E segmented pixels have Delta L*, delta a* and Delta b* colour values. Delta L stands for lightness difference among the anterior ciliary artery and conjunctiva. Delta a* is the redness difference between the anterior ciliary artery and conjunctiva. Delta b* is the blueness-yellowness difference between the anterior ciliary artery and conjunctiva. Total colour difference (Delta E*) obtained from delta L*, a*, b* colour difference shows the distance of a line between anterior ciliary artery and conjunctiva. Delta E segmentation shows between two colour regions: ciliary artery and conjunctiva or sclera. The RGB image is converted to LAB colour space in delta segmentation of the ciliary artery. The user can select an irregular shaped region by freehand drawing to identify the colour of the anterior ciliary artery. To extract the original image’s anterior ciliary artery region colour bands, separate them into three 2D arrays or channels as shown in
Person |
Gender/Age/weight |
Hg (g/L) |
Frequency count & |
Frequency count & |
---|---|---|---|---|
I | M/21/55 | 9.2 | (1400-1800) & (5-8) | (1400-1700) & (5-7) |
II | F/32/67 | 12.4 | (1400-1700) & (5-10) | (1400-1600) & (5-8) |
III | M/36/48 | 10.8 | (1400-1900) & (5-9) | (1400-1800) & (5-8) |
IV | F/12/69 | 14.2 | (1400-2500) & (5-11) | (1400-2300) & (5-9) |
V | M/34/77 | 15.4 | (1400-3000) & (5-12) | (1400-2900) & (5-11) |
VI | M/55/61 | 11.7 | (1400-1600) & (5-11) | (1400-1500) & (5-10) |
VII | F/42/82 | 8.5 | (1400-1700) & (5-7) | (1400-1600) & (5-7) |
Each channel represents a colour component and calculates the average value of lab colour. The delta E image represents the colour difference. The difference in anterior ciliary artery region colour in LAB colour space (Delta E) is computed for each pixel in the image between that pixel’s colour and the average LAB colour of the anterior ciliary artery region drawn. The delta E is the square root of the sum of the squares of the delta anterior ciliary artery region images. The value of delta E in the mask anterior ciliary artery region gives the mean delta E value. Calculate the value of the standard deviation of the delta E. The proposed SBH method predicts the haemoglobin from the anterior ciliary artery region using average mean value and intensity count from the image and laboratory value through a Gaussian regression model.
Anterior Ciliary Artery Region Segmented from Image | TDyWT& |
TDyWT& |
TDyWT& |
TDyWT& |
||||||
---|---|---|---|---|---|---|---|---|---|---|
Right Eye | Right Eye | Left |
Left |
Right |
Right |
Left Eye | Left Eye | |||
Statistical parameters | ||||||||||
Mean | 12.0803 | 10.2448 | 5.7257 | 5.3155 | 40.8525 | 39.6381 | 67.6142 | 65.1869 | ||
Standard Deviation | 43.5893 | 41.7981 | 30.2457 | 32.6328 | 65.3014 | 65.5710 | 78.0856 | 75.2145 | ||
Entropy | 0.8658 | 1.1059 | 0.4645 | 0.4813 | 3.2736 | 2.9661 | 4.7428 | 3.1303 |
Gaussian process regression (GPR) predicts haemoglobin through the mean value of the eye, intensity range, frequency count and laboratory value. GPR is non-parametric performs well for small datasets, and uncertainty prediction-based measurement is obtained. GPR is a tuned version of the Bayesian method. In linear function y = wx+ £, p (w) is the prior distribution on parameter w, and the Bayes rule is applied. The predictive posterior distribution for GPR is given by
New |
Haemoglobin (proposed SBH |
Haemoglobin |
Conjunctiva region/ accuracy percentage (Borescope image) |
---|---|---|---|
NP-I | 12.9/94% | 13.7 | 11.8/86% |
NP-II | 8.5/95% | 8.9 | 7.1/80% |
NP-III | 11.2/96% | 11.8 | 10.2/87% |
NP-IV | 14.3/95% | 15.1 | 13.5/89% |
NP-V | 15.7/96% | 16.2 | 14.5/90% |
Repetitive measurement of haemoglobin for pulmonary hypertension patients and pregnant women help clinicians for analyzing the swing in haemoglobin values. Repetitive measurement through invasive haemoglobin measurement leads to bacterial infections. For avoiding invasive measurement of haemoglobin, the noninvasive measurement through anterior ciliary arteries region pixels obtained from borescope camera attached smartphone is proposed. The proposed SBH method has more accuracy in haemoglobin measurement than the conjunctiva region due to measurement from red blood cells in the blood from ciliary arteries of the eye. The problem of a decrease in the density of goblet cells and acinar units in the conjunctiva region-based haemoglobin measurement leads to inaccuracy. The above problem-solve by measuring haemoglobin levels from ciliary arteries of left and right eye red blood cells. Red blood cells are initially projected in the anterior ciliary artery region and then delta segmented to extract only red blood cells. Perspective projection of the anterior ciliary artery region reduces the noise pixels from the sclera region of the eye.
Moreover, the frequency count and intensity values of red cells of the artery are considered for the haemoglobin measurement through the Gaussian regression method. The red blood cells average values from the image, frequency count and intensity values, age, and weight-based measurement leads to more accuracy in predicting the haemoglobin in human. Furthermore, anterior ciliary artery region-based Polycythemia Vera condition can be predicted, which causes a blood clot.