--Wave2-pics-July-05-2022 contained labeled images of patients (face, home, kitchen, and toilet) during  wave 1 home 
health visits.

--Labeling is done manaually by linking the qualtriccs survey ID number with the image numbers (images are also collected 
in the qualtrice survey as file uploads)

--Labels have 3 layers, "notwell", "ok" and "good", indicating their self report health status (It is based on the EQ-5D questions on patients' mobility, physical well-being, emotional well-beinig and their overall score of utility.)

-- The wave 2 completed questionnaires is attached as excel sheet, it contains 110 variables that cover patients demographics, income, 
health utilization, health cost, as well as blood pressure, blood glocuse, weights, hegihts. 



Ways to use the data
Use iamges + other predictors to predict the health risk of a patient. 
Binary predictor: High Risk (need immediate medical attention from a nurse/doctor) and Low Risk

Possible sources we can use:


Tensorflow (cats and dogs classification example)
https://www.tensorflow.org/datasets/catalog/cats_vs_dogs

Tensorflow for web and mobile development
https://www.tensorflow.org/install


PyHealth:
https://github.com/zzachw/PyHealth

Theory: 
https://ieeexplore.ieee.org/document/554195
https://arxiv.org/abs/1409.1556
https://arxiv.org/pdf/1704.04861.pdf