The blood hemoglobin level (Hgb) measurement has a vital role in the diagnosis, evaluation, and management of numerous diseases. This paper describes the use of a smartphone video imaging and an artificial neural network (ANN) system to estimate Hgb levels non-invasively. We collected 10 second- (300 frame) captured fingertip videos using smartphone in 75 adults. Red, green, and blue pixel intensity were estimated for each of 100 area blocks in each frame and the patterns across the 300 frames were described. ANN was then used to develop a model using the extracted video features to predict hemoglobin levels. In our study sample, with patients 20-56 years of age, and gold standard hemoglobin levels of 7.6 to 13.5 g/dL., we observed a 0.93 rank order of correlation between model and gold standard hemoglobin levels. Moreover, we identified specific regions of interest in the video images which reduced the required feature space.

Learning Objective 1: Potential of exploiting machine learning algorithms in identifying hemoglobin level in non-invasive manner

Learning Objective 2: Importance and mechanism of reducing feature space without sacrificing the accuracy level


Md Hasan (Presenter)
Marquette University

Md Haque, Purdue University
Riddhiman Adib, Marquette University
Jannatul Ferdause Tumpa, Marquette University
Richard Love, Marquette University
Azima Begum, Dhaka Medical College and Hospital
Young Kim, Purdue University
Sheikh Ahamed, Marquette University

Presentation Materials: