Measuring hemoglobin levels by taking pictures of fingernail beds with a smartphone

The Sanguina Smartphone App measures hemoglobin levels using pictures of fingernail beds. Our study tests this App in a real world, low resource setting, in rural Bihar, India. This post describes the rationale behind this study and summarizes its results.
Published in Healthcare & Nursing
Measuring hemoglobin levels by taking pictures of fingernail beds with a smartphone
Like

Share this post

Choose a social network to share with, or copy the URL to share elsewhere

This is a representation of how your post may appear on social media. The actual post will vary between social networks

Anemia constitutes a serious global health problem. Through its adverse effects on human health, it has a substantial impact on socioeconomic development. In Bihar, India, 68 percent of children are affected by anemia [1]. Iron-deficiency is one common cause of anemia. We aimed to contribute to the fight against anemia by introducing a method of iron supplementation to pre-schools, so-called Anganwadi Centres, in Bihar. In order to evaluate the effectiveness of the intervention, it was necessary to measure hemoglobin levels among children attending Anganwadi Centres.

In previous studies in this setting, we had worked with the HemoCue Hb 301 to measure hemoglobin. This device is a handheld photometer which is commonly used in field-based settings and is considered as reference standard in a setting where no lab access is available. It requires to draw a drop of blood using a finger prick. As we planned to measure hemoglobin of several thousand children, we searched for alternative, non-invasive methods and came across the Sanguina Smartphone App. This App claimed to measure hemoglobin based on an extremely color-sensitive analysis algorithm which evaluates the image data of photos taken from fingernail beds. The initial validation study, conducted in a laboratory setting in the US, reported an accuracy of ±2.4 g/dl [2]. While the App was not yet publicly available, its developers agreed to provide the App to us for validation in our specific study setting. Results of the validation study are presented in the linked paper and summarized in this post. 

For our validation study, we measured hemoglobin levels in two different samples, using the Smartphone App and a reference method. The first sample consisted of mainly adult patients at a community-based clinic in Madhepura city. These patients had venous blood drawn for analysis after a doctor’s visit. This allowed us to measure hemoglobin via a complete blood count using the best possible local standard, the Aspen Mindray BC-5000 autoanalyzer. This method served as the reference value for the validation of the App. The second sample consisted of children aged 2 to 7 years attending five Anganwadi Centres in remote villages in Madhepura district. This captured the population group we were interested in for the iron supplementation intervention. It was not possible to draw venous blood from the children and do a complete blood count evaluation in this remote setting or transport blood samples to the nearest laboratory. We therefore chose the commonly used point-of-care device, the HemoCue Hb 301, to determine the reference hemoglobin value in this sample. For both samples, we compared the respective reference measurements with those obtained with the App.

First results indicated that the App was less precise than we had hoped. For the clinic-based sample, we found an accuracy of ±4.43 g/dl, while for the pre-school sample, the accuracy was ±3.54 g/dl. This was much less accurate than the ±2.4 g/dl found in the initial study in the US laboratory.

While the App did not seem to capture hemoglobin levels accurately, repeatability observed between successive measurements was high. This means that the App gave consistent results for different pictures of the same individual’s hands. We therefore hypothesized that the App was not used to the types of nails prevalent in our study setting, as it had been calibrated in a US laboratory. Fingernail discolorations, e.g. due to medical conditions, injuries, remnants of henna or nail polish, were prevalent and led to measurement inaccuracies of the extremely color-sensitive algorithm of the App. These thoughts led to two further analyses.

Firstly, we assessed the relationship between nail quality and accuracy of the App. All nails were categorized according to their quality based on visual inspection of the pictures. Analyses were re-run for nails falling into the two highest categories. Accuracy did not greatly improve. In the clinic-based sample, accuracy was ±4.19 g/dl, while in the pre-school sample, it was ±3.00 g/dl.

Secondly, we collaborated with the developers of the App to re-train the algorithm using pictures collected in the clinic-based sample. For this re-training, only pictures of nails in the highest category, with least discolorations, were used. The re-trained algorithm was then used to re-evaluate the pictures of the pre-school sample and we compared the newly measured hemoglobin levels with those measured by the reference method. This re-training led to a considerable improvement of the accuracy of the App. In the pre-school sample, accuracy was ±2.25 g/dl – very close to the accuracy measured in the initial study in the US laboratory.

From our findings, we drew several conclusions. Our study showed the ability of the App to be adapted to a different context and population. With further improvements, the App could therefore become a viable tool in low-resource point-of-care settings. In such settings, the alternative to a slightly inaccurate measurement would likely be no measurement due to resource constraints. As the use of the App is easily trained and smartphones become more widespread as a work tool among community healthcare workers, it could be a scalable solution to measure hemoglobin and refer potential anemia cases for further testing.

Moreover, the App could be very suitable for use in quantitative surveys. Using the App for measuring hemoglobin levels is easier, less costly, less wasteful, less risky, and less painful than using invasive methods such as the HemoCue. Enumerators can easily be trained in its use, even if they are not trained nurses.

Finally, validation of measurement tools in different settings is crucial. Relying on results of the initial study, which was conducted in a very different setting from ours, could have led to very misleading results when evaluating the iron supplementation intervention.

 

References

[1]       International Institute for Population Sciences (IIPS) and ICF, ‘National Family Health Survey (NFHS-5), 2019-21: State Fact Sheet Bihar’, Mumbai, 2021.

[2]       R. G. Mannino et al., ‘Smartphone app for non-invasive detection of anemia using only patient-sourced photos’, Nat. Commun., vol. 9, no. 1, Art. no. 1, Dec. 2018, doi: 10.1038/s41467-018-07262-2.

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Follow the Topic

Health Care
Life Sciences > Health Sciences > Health Care

Related Collections

With collections, you can get published faster and increase your visibility.

Applications of Artificial Intelligence in Cancer

In this cross-journal collection between Nature Communications, npj Digital Medicine, npj Precision Oncology, Communications Medicine, Communications Biology, and Scientific Reports, we invite submissions with a focus on artificial intelligence in cancer.

Publishing Model: Open Access

Deadline: Dec 30, 2024

Health in Africa

We aim to promote high-quality research that advances our understanding of health issues in Africa, and advocates for better healthcare on the continent in line with the UN’s SDGs.

Publishing Model: Open Access

Deadline: Dec 31, 2024