论文阅读和分析:Applying a Deep Learning Network in Continuous Physiological Parameter Estimation Based on Photoplethysmography Sensor Signals
## 主要内容:
1、使用CNN-LSTM神经网络架构同时计算HR、SBP、DBP、MAP(心率和血压);
2、通过Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS)的标准;
3、在MIMIC II数据集上使用10-fold交叉验证;
4、采样频率125HZ;
5、预处理,将数据集中的数据长度过短、数据值异常的去掉;
网络架构:
网络层配置:
损失函数:
H
δ
(
y
,
f
(
x
)
)
=
{
1
2
(
y
−
f
(
x
)
)
2
,
i
f
∣
y
−
f
(
x
)
∣
≤
δ
,
δ
∣
y
−
f
(
x
)
∣
−
1
2
δ
2
,
o
t
h
e
r
w
i
s
e
.
H_\delta(y,f(x))=\begin{cases}\frac{1}{2}(y-f(x))^2,&if~|y-f(x)|\le\delta,\\[6pt]\delta|y-f(x)|-\frac{1}{2}\delta^2,&otherwise.\end{cases}
Hδ(y,f(x))=⎩
⎨
⎧21(y−f(x))2,δ∣y−f(x)∣−21δ2,if ∣y−f(x)∣≤δ,otherwise.
其中
δ
=
1
\delta=1
δ=1
评价指标:
M A E = 1 n ∑ i = 1 n ∣ y i − y ^ i ∣ M E = 1 n ∑ i = 1 n ( y i − y ^ i ) S D = 1 n ∑ i = 1 n ( x i − M E ) 2 \begin{aligned} {MAE}& =\frac{1}{n}\sum\limits_{i=1}^n\left|y_i-\hat{y}_i\right| \\ ME& =\frac{1}{n}\sum\limits_{i=1}^n\left(y_i-\hat{y}_i\right) \\ SD& =\sqrt{\frac{1}{n}\sum\limits_{i=1}^n\left(x_i-ME\right)^2} \end{aligned} MAEMESD=n1i=1∑n∣yi−y^i∣=n1i=1∑n(yi−y^i)=n1i=1∑n(xi−ME)2
实验结果:
参考:
Applying a Deep Learning Network in Continuous Physiological Parameter Estimation Based on Photoplethysmography Sensor Signals