Many diseases can be found by looking at the retina?

A few days ago, the "Nature" sub-Journal "Nature-Medical Engineering" published a new paper on the use of machine learning technology for disease diagnosis

This research was brought by the joint team of Professor Zhang Kang, Professor Zhou Yong, Professor Wang Guangyu, Professor Xu Tao, and Professor Chen Ting

Why make a fuss on the fundus of the retina? This is because the retina may contain many harbingers of diseases-these diseases will affect the capillaries, nerves and connecting tissues of the retina, which reflect abnormalities in the structure

In the past few years, the use of machine learning technology for medical image processing and disease diagnosis has been an emerging field

▲Model design of this research (picture source: reference [1])

And this is the meaning of this thesis

Specifically, the researchers first established a series of models for the detection of chronic kidney disease and type 2 diabetes: some of these models are based on other clinical data of the patient (such as age, gender, height, blood pressure, etc.

The comparison results show that, in the diagnosis of chronic kidney disease, the AUC value of the former (representing the accuracy of the model in judging "disease" and "no disease", the limit is 1) is 0.

▲In the diagnosis of chronic kidney disease, the model (orange) performed better than the model (green) generated from clinical data

Similarly, researchers have observed the same trend in diagnosing type 2 diabetes

It is worth mentioning that in the past, researchers have built similar models, but they have not achieved the desired results in external data verification-models based on patient clinical data perform better, while models using retinal imaging will slow down.

On the contrary, in this study, the researchers used multiple external data for verification, and the data covered different ethnic groups and populations, and all achieved consistent results: the retinal imaging model is better than the model using only clinical data, and the combination of the two To achieve the best results

▲Analysis of different ethnic groups and populations has confirmed the reliability of the model (picture source: reference [2])

In the paper, the researchers also evaluated some other application possibilities, such as whether it is possible to use a smartphone to take pictures of the retina for analysis, or whether it can predict who is more likely to develop chronic kidney disease and type 2 diabetes in the future

The review of a special article in "Nature-Medical Engineering" pointed out that although this model cannot be directly used for direct diagnosis of patients, it is expected to play its strengths in population-scale disease screening projects

Reference materials:

[1] Zhang, K.

[2] Mitani, A.

Original title: Can you find multiple diseases by looking at the retina? The Chinese team's "Nature" sub-news introduces new diagnostic tools