This project has the intention to propose machine learning techniques in temporal series for Photoplethysmogram data obtained by smarthpones and avaluate its suitability comparing it with tradicional exams for measurement of hemoglobin, bilirubin, oxygen saturation and leukocytes. First, data from voluntary pacients of the UFSCar's Univesitary Hospital will be obtained, they will give their blood test result, as well as a video of their finger filmed with a smartphone in the "flash on" mode, and a photo of their eyes. Then, video processing techniques will be used in the videos to the genaration of time series, that will be used in the creation on regression models for the estimation of the clincal parameters. In the creation of the model, Recurrent Neural Networks (RNNs) will be used and traind with the obtained time series and annotated from laboratory test results. In case of success, the model could be used in smartphones to the prediction of those clinical parameters, without the need of expensive and invasive processes.
News published in Agência FAPESP Newsletter about the scholarship: