TY - JOUR
T1 - Non-Invasive Estimation of Hemoglobin Using a Multi-Model Stacking Regressor
AU - Acharya, Soumyadipta
AU - Swaminathan, Dhivya
AU - Das, Sreetama
AU - Kansara, Krity
AU - Chakraborty, Sushovan
AU - Kumar R, Dinesh
AU - Francis, Tony
AU - Aatre, Kiran R.
N1 - Funding Information:
Manuscript received January 18, 2019; revised July 3, 2019 and September 10, 2019; accepted November 10, 2019. Date of publication November 20, 2019; date of current version June 5, 2020. This work was supported in part by USAID Saving Lives at Birth Grand Challenges for funding, in part by CIMIT’s Primary Health Care Challenge award (second place), and in part by the Robert Bosch Engineering and Business Solutions Pvt. Ltd. (RBEI). (Corresponding author: Soumyadipta Acharya.) S. Acharya is with Johns Hopkins University, Baltimore, MD 21218 USA (e-mail: acharya@jhu.edu).
Publisher Copyright:
© 2013 IEEE.
PY - 2020/6
Y1 - 2020/6
N2 - Objective: We describe a novel machine-learning based method to estimate total Hemoglobin (Hb) using photoplethysmograms (PPGs) acquired non-invasively. Methods: In a study conducted in Karnataka, India, 1583 women (pregnant and non-pregnant) of childbearing age, with Hb values ranging between 1.6 to 14.8 g/dL, had their Hb values estimated using intravenous blood samples and concurrently by a finger sensor custom designed and prototyped for this study. The finger sensor collected PPG signals at four wavelengths: 590 nm, 660 nm, 810 nm, and 940 nm. A novel feature vector was derived from these PPGs. A machine learning model comprising of a two-layer stack of regressors including Least Absolute Shrinkage and Selection Operator (LASSO), Ridge, Elastic Net, Adaptive (Ada) Boost and Support Vector Regressors (SVR) was designed and tested. Results: We report a statistically significant Pearson's correlation coefficient (PCC) of 0.81 (p < 0.01) between the Hb value estimated by the proposed methodology and gold standard values of Hb, with a Root Mean Square Error (RMSE) of 1.353 ± 0.042 g/dL. The performance of the stacked regressor model was significantly better than the performance of individual regressors (low RMSE, and better CC; p < 0.05). Post-hoc analysis showed that including pregnant women in the training data set significantly improved the performance of the algorithm. Conclusion: This article demonstrates the feasibility of a machine learning based non-invasive hemoglobin measurement system, especially for maternal anemia detection. Significance: By developing and demonstrating a machine learning approach on a large data set, we have demonstrated that such an approach could become the basis for a public health screening tool to detect and treat maternal anemia and could supplement global health intervention strategies.
AB - Objective: We describe a novel machine-learning based method to estimate total Hemoglobin (Hb) using photoplethysmograms (PPGs) acquired non-invasively. Methods: In a study conducted in Karnataka, India, 1583 women (pregnant and non-pregnant) of childbearing age, with Hb values ranging between 1.6 to 14.8 g/dL, had their Hb values estimated using intravenous blood samples and concurrently by a finger sensor custom designed and prototyped for this study. The finger sensor collected PPG signals at four wavelengths: 590 nm, 660 nm, 810 nm, and 940 nm. A novel feature vector was derived from these PPGs. A machine learning model comprising of a two-layer stack of regressors including Least Absolute Shrinkage and Selection Operator (LASSO), Ridge, Elastic Net, Adaptive (Ada) Boost and Support Vector Regressors (SVR) was designed and tested. Results: We report a statistically significant Pearson's correlation coefficient (PCC) of 0.81 (p < 0.01) between the Hb value estimated by the proposed methodology and gold standard values of Hb, with a Root Mean Square Error (RMSE) of 1.353 ± 0.042 g/dL. The performance of the stacked regressor model was significantly better than the performance of individual regressors (low RMSE, and better CC; p < 0.05). Post-hoc analysis showed that including pregnant women in the training data set significantly improved the performance of the algorithm. Conclusion: This article demonstrates the feasibility of a machine learning based non-invasive hemoglobin measurement system, especially for maternal anemia detection. Significance: By developing and demonstrating a machine learning approach on a large data set, we have demonstrated that such an approach could become the basis for a public health screening tool to detect and treat maternal anemia and could supplement global health intervention strategies.
KW - Anemia
KW - Beer-Lambert's law
KW - machine learning
KW - non-invasive hemoglobin
KW - plethysmography
KW - stacked regressor
UR - http://www.scopus.com/inward/record.url?scp=85086267656&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85086267656&partnerID=8YFLogxK
U2 - 10.1109/JBHI.2019.2954553
DO - 10.1109/JBHI.2019.2954553
M3 - Article
C2 - 31751256
AN - SCOPUS:85086267656
SN - 2168-2194
VL - 24
SP - 1717
EP - 1726
JO - IEEE Journal of Biomedical and Health Informatics
JF - IEEE Journal of Biomedical and Health Informatics
IS - 6
M1 - 8907371
ER -