Deep learning-detect diabetes with ECG

Thesis title

Deepheart:semi-supervisedsequencelearningforcardiovascularrisk Prediction

Recommended Index: * * * * *
Recommendation reason: The idea is very new, discovered the human body signal Some novel association

A word summarizing the main things of this paper:

Use the heart rate data from the bracelet to detect four diseases: diabetes, high cholesterol, sleep apnea and high blood pressure

The company's main starting point:

Found that these four kinds of diseases are very difficult to be aware of their own, so with some non-exclusive medical equipment to do early warning, to help people early detection of their illness.

Target Users : general public, non-sick users

The most ingenious part of the article is

Found a simple bracelet data somehow can detect a number of other diseases
(a bit similar to the pulse diagnosis is that the human body is an interrelated whole, the state of the body will be reflected in the pulse)

Data

57,675 Week data from 14022 people

Collaborating with the University of California's cardiology department.

Label Source: Participants were previously diagnosed with the body

Main model

Supervisory Section : one-dimensional convolution plus lstm

preprocessing Section :

1. Semi-supervised training with a autoencoder (three-ply convolution + 4-layer cyclic layer)
2. Heuristic training is to pre-train the variable of small time window as target.

Summarize

The paper found a very interesting entry point, is a start-up paper, and is currently developing a product based on this paper, is a very meaningful thing, can help people in the case of new hardware without the need to detect the risk of disease in advance, than those bored purely for the paper to do the deep learning application is much better.

Deep learning-detect diabetes with ECG