Page 71 - 22-0424

The International Journal of the Royal Society of Thailand
Volume XI - 2019

neural networks are convolutional neural networks and recurrent neural networks
(Cheng et al., 2016; Kellum and Bihorac, 2019). Convolutional neural networks
consist of additional layer of neurons that connected only to some neurons in the

previous layers. This adding layers are part of deep learning reducing resource
requirement in data processing. Recurrent neural networks are different from
feedforward neural network in that they use their past output as new input to
form feedback loops. Therefore, they were designed to deal with sequentially
ordered data such as time series and location series (Spoerer et al., 2017; Yu et al.,
2019).

Big data and AI can be used to improve all aspects of AKI care process
(Figure1). In patients without AKI, AI can be used to detect patients with high
risk of AKI and early alert physicians. Moreover, in AKI patients, AI can be used
to select the best therapy options to minimize complications. In addition, after
AKI, AI can be used to predict risk of AKI to CKD transition. We listed the recent
studies of AI in AKI in Table 2.

Table 2 List of studies using AI in acute kidney injury.

Article Year Type of AI Input N Outcome

Prediction 2018 Gradient Adult In 121,158 Predict stage 2 AKI
of AKI Boosting patients in 24hours with
Koyner et al. Machine University of AUC 0.90 and pre-
(Koyner et al., (GBM) Chicago dict stage 2 AKI in
2018) 48 hours with AUC
0.87

Tomašev et al. 2019 Recurrent Medical record 703,782 Predict AKI in 48
(Tomašev neural network from US hours; able to ob-
et al., 2019) (RNN) Department tained 55% of AKI
of Veterans patients with 2:1
Affairs false alert to true
alert ratio

Tranet et al. 2019 k-nearest neigh- 20% or more 50 Predict AKI with
(Tran et al., bors algorithm surface area use of biomarker
2019) burn patients and AI with accu-
racy more than 90%

Phatadon Sirivongrangson
Nattachai Srisawat 65

11/7/2565 BE 13:28
_22-0424(055-076)7.indd 65 11/7/2565 BE 13:28
_22-0424(055-076)7.indd 65