In a recent study published in the PLOS ONE journal, researchers assessed the feasibility of smartphone facial colorimetry as an anemia screening technique.

Background

Taking into account disability-adjusted life years, anemia is the 14th leading cause of disease worldwide. Developmental outcomes can be negatively impacted by pediatric anemia, while iron deficiency anemia is related to a lower cognitive development index, which often improves substantially after therapy. Iron deficiency may result in long-lasting alterations in neuronal physiology and biochemistry if it is not treated promptly. Given the obvious advantages of early treatment and diagnosis, it is essential to comprehend the local and global variables that contribute to its absence.

About the study

In the present study, researchers employed smartphone-based colorimetry to provide a non-invasive method for screening for anemia in newborns and young children.

A convenience sample of 62 children aged from birth to 59 months was recruited from February to April 2018 at the Korle Bu Teaching Hospital in Ghana. Included participants were clinically stable, had a blood test was ordered as part of their usual clinical care, and whose urgent care would not be postponed by image acquisition in this study. Blood hemoglobin concentration was tested for each participant using a HemoCue Hb 301 anemia screening equipment.

The imaging was performed utilizing the rear-facing camera of a Samsung Galaxy S8 smartphone in conjunction with a proprietary mobile application that collected and stored lossless and raw camera photos. The use of raw pictures indicates minimal post-sensor processing. Three sets of photos were captured, including images of the sclera and surrounding skin, the folded-over lower eyelid, and the lower lip. The team manually segmented three regions of interest (ROIs) for each participant, namely, the sclera, the lower lip, and the lower palpebral conjunctiva. The photos were then pre-processed to adjust for different ambient illumination conditions, after which blood hemoglobin concentration measurements were collected.

After adjusting the image for ambient illumination, potential color measurements for each ROI were determined. Accordingly, five separate features classes were measured from the preprocessed image, including the erythema index, g chromaticity, r chromaticity, a* chromaticity, and b* chromaticity. For each feature and ROI, the team extracted the median as well as the median of the top 5% of the features.

Results

The average age of the participants was almost 1.25 years. The average hemoglobin concentration detected in the blood was approximately 11.7 g/dL. For 19 of the eligible individuals, pictures of certain ROIs were not obtained or needed to be of adequate quality. The remaining patient photos had an average subtracted signal-to-noise ratio (SSNR) of 5.70. After low-quality photos were omitted, one-way analysis of variance (ANOVA) revealed that the SSNR varied significantly between the various ROIs.

The correlation coefficient of the model was strongly impacted by choice of redness estimate and ROI, but not by the subtraction technique. The optimal redness measurement was g chromaticity while the optimal area of focus was the sclera. Almost 15 out of 270 predictors displayed statistically significant associations with the evaluated blood hemoglobin concentration.

The optimal combination of pipelines containing one feature from each ROI was chosen for the multi-ROi model. In this case, the lip feature was now the median of the top 5% of r chromaticities, with ambient subtraction accounting for the top 5% of pixels by redness. After rectification with white balancing in the flash image, the eyelid feature was the median of the top 5% of g chromaticities, with the white point chromaticity captured from the 15% of pixels on the sclera region with the lowest redness.

Conclusion

The study findings showed that white balancing utilizing a reference point on the sclera functioned well while analyzing the sclera, whereas ambient subtraction was more effective elsewhere. Depending on the ROI, the best redness metric for predicting blood hemoglobin content was either the erythema index, r chromaticity, or the CIE 1976 b* chromaticity.

The best algorithm achieved 92.9% sensitivity and 89.7% specificity while screening for anemia, in accordance with World Health Organization recommendations. This adds to the growing body of evidence indicating that smartphones may be effective for anemia screening. Before such methodologies can be applied to medical devices, however, obstacles pertaining to the inclusion of diverse populations as well as the presentation of results must be overcome.

Written by

Bhavana Kunkalikar

Bhavana Kunkalikar is a medical writer based in Goa, India. Her academic background is in Pharmaceutical sciences and she holds a Bachelor's degree in Pharmacy. Her educational background allowed her to foster an interest in anatomical and physiological sciences. Her college project work based on ‘The manifestations and causes of sickle cell anemia’ formed the stepping stone to a life-long fascination with human pathophysiology.