Gauging blood hemoglobin with a photograph

Blood hemoglobin tests are routinely administered to patients who have conditions such as anemia, kidney disease, and cancer or are hemorrhaging after a traumatic injury. Portable, point-of-care blood analyzers can measure hemoglobin from drawn blood, but they rely on environment-sensitive cartridges with short shelf lives. They are also unaffordable for resource-limited and home care settings.

Purdue University doctoral candidate Sang Mok Park, his adviser Young Kim , and their colleagues have now developed a new mobile-health smartphone application that estimates a patient’s blood hemoglobin level. The procedure could hardly be less invasive: Instead of blood being drawn, a photograph is taken of the patient’s inner eyelid. The microvasculature there is easily visible and unaffected by skin color, so a patient need only lower an eyelid to expose the tissue, as shown above. Software currently running on a separate computer essentially transforms the smartphone’s built-in camera into a virtual spectral imager.

The software collects low-resolution red, green, and blue (RGB) data from the image and applies a least-squares regression algorithm to mathematically reconstruct a much richer set of multiwavelength data. To pull off the conversion, the researchers needed a data set of blood hemoglobin levels from real patients. They got them from a clinical study of 153 participants at Moi Teaching and Referral Hospital in Eldoret, Kenya. Eyelid photographs were taken from participants immediately before or after their blood was drawn and processed via a commercial hematology analyzer. A portion of the data was used to train the algorithm to convert the RGB data into high-resolution spectra. Because the spectral profile reflected from the inner eyelid is sensitive to changes in the hemoglobin content in the blood, the reconstructed eyelid spectrum could then be processed by the software to precisely predict the hemoglobin concentration.

Tests of other participants in the Moi study confirmed that the application’s predictions of a patient’s hemoglobin counts agreed with standard laboratory measurements to within 5–10%. According to the researchers, the software can be extended to different smartphone models by incorporating the spectral response functions in the R, G, and B channels of the camera. (S. M. Park et al., Optica 7, 563, 2020 .)