Processing method and device of hematoma aspiration decision-making system for intracerebral hemorrhage

1 . A processing method of a hematoma aspiration decision-making system implemented in and physically controlling a hematoma aspiration apparatus comprising a micro flowmeter, a gravimeter, a sphygmomanometer, an oximeter, a heart rate monitor, an intracranial pressure monitor, a near-infrared measuring instrument, a near-infrared probe and a micro metering pump during an intracerebral hemorrhage surgery, wherein the method comprises:

acquiring preoperative data for hematoma aspiration, wherein the preoperative data for hematoma aspiration involves a first indicator;

inputting the preoperative data for hematoma aspiration into a first deep neural network to determine an adapted initial hematoma aspiration protocol preoperatively, wherein the first deep neural network is configured to perform multi-center clustering on the preoperative data for hematoma aspiration input into the network and output a hematoma aspiration protocol preadapted by a most adapted clustering center;

acquiring real-time intraoperative data for hematoma aspiration through the hematoma aspiration apparatus, wherein the real-time intraoperative data for hematoma aspiration involves a second indicator, the second indicator comprises an aspiration velocity, a fluid inlet velocity, blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, and a hematoma size;

selecting, from a plurality of pre-trained Long Short-Term Memory (LSTM) networks, a specific LSTM network that is preadapted by and corresponds to the most adapted clustering center determined by the first deep neural network;

inputting the real-time intraoperative data for hematoma aspiration into a second deep neural network distinct from the first deep neural network, wherein the second deep neural network corresponds to the selected specific LSTM network, so as to determine a decision-making result of the hematoma aspiration protocol and a prediction result of the hematoma aspiration protocol intraoperatively, wherein the second deep neural network is configured to determine, based on the real-time intraoperative data for hematoma aspiration input into the network, a time series control protocol corresponding to a real-time hematoma aspiration process, and the time series control protocol comprises protocol content of the hematoma aspiration protocol in a decision-making aspect and a prediction aspect, wherein the decision-making result comprises control amounts for a fluid inlet amount and a fluid outlet amount by the micro metering pump, and the prediction result comprises a prediction result amount and a prediction decision amount, wherein the prediction result amount comprises blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, and a hematoma size, and the prediction decision amount comprises the fluid inlet amount and the fluid outlet amount; and

generating, based on the decision-making result, one or more control signals that directly drive the micro metering pump of the hematoma aspiration apparatus to dynamically and physically control a fluid inlet amount through the fluid inlet channel and a fluid outlet amount through the fluid outlet channel during the intracerebral hemorrhage surgery, wherein the micro flowmeter provides instantaneous flow-rate feedback, the gravimeter measures mass change to correct the flow-rate measurement; wherein the method implements a real-time closed-loop control of the hematoma aspiration apparatus by iteratively updating the hematoma aspiration protocol based on newly acquired intraoperative data at each subsequent moment, thereby achieving real-time and high-accuracy control of the hematoma aspiration process;

for the second deep neural network, the time series control protocol being represented as:

{

H

1

,

H

2

,

…

,

H

t

,

…

,

H

T

}

,

H t is a control protocol at a moment t, T is a maximum value of t, and T corresponds to a total time,

H

t

=

{

C

t

,

S

t

,

D

t

}

,

 wherein

H t comprises a number of states at the moment t, comprising a control state C t , a static state S t , and a dynamic state D t , specifically:

C t serves to control an action at a next moment of the hematoma aspiration device, comprising a fluid inlet type, a fluid inlet velocity, and an aspiration velocity,

S t is a descriptor value rather than a specific numerical value, serves as state monitoring, and comprises a heart rate, blood pressure, and blood oxygen,

D t is used to compare with an actual value to correct a control protocol H t+1 at the next moment, comprising intracranial pressure, a hematoma size, and a hematoma location,

the control protocol H t+1 at the next moment is calculated by the following equation:

H

t

+

1

=

δ

[

ω

i

(

H

t

,

x

t

)

+

b

i

]

,

 wherein

X t is an actual state at the moment t, comprising intracranial pressure, a hematoma size, and a hematoma location that are actually measured, δ, ω i , and b i are different network parameters of the second deep neural network, and x t , and D t are used to calculate a decision error and correct a decision value at the next moment by using the decision error so as to obtain an actual decision.

2 . The method according to claim 1 , wherein the first indicator involved in the preoperative data for hematoma aspiration specifically comprises:

an age, an NIHSS score, a medical history, blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, a hematoma volume, a CT image, and an MRI image; and

the second indicator involved in the intraoperative data for hematoma aspiration specifically comprises:

an aspiration velocity, a fluid inlet velocity, blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, and a hematoma size.

3 . The method according to claim 1 , wherein the protocol content of the initial hematoma aspiration protocol comprises:

a tolerance range of changes in intracranial pressure, a single aspiration duration, a hematoma state, whether other drug injections are needed, and a maximum aspiration velocity.

4 . The method according to claim 1 , wherein the method further comprises:

displaying H t , C t , S t , D t , X t , and the decision error through a visual interface;

receiving a control action Y t input by a decision-making participant, wherein Y t comprises at least one of changing the fluid inlet type, adjusting the fluid inlet velocity, and adjusting the aspiration velocity; and

executing C t and Y t , and replacing H t with H′ t , wherein in H′ t , C′ t =C t +Y t , a remaining part is the same as H t , and a calculation mode of H t+1 is replaced with:

H

t

+

1

=

δ

[

ω

i

(

H

t

′

,

x

t

)

+

b

i

]

,

H

t

′

=

{

C

t

′

,

S

t

,

D

t

}

,

C

t

′

=

C

t

+

Y

t

.

5 . The method according to claim 4 , wherein the method further comprises:

standardizing eigenvalues obtained by numeralization of H t , C t , S t , D t , X t , and Y t , and recording as a decision-making matrix represented as follows:

x

0

x

1

…

x

t

C

0

C

1

…

C

t

S

0

S

1

…

S

t

D

0

D

1

…

D

t

Y

0

Y

1

…

Y

t

,

in the decision-making matrix, each column represents all states and decision eigenvalues at a certain moment, and each row represents values of a certain feature at all moments;

calculating an error between the decision-making matrix and an actual perceived value by the following equation:

error

=



h

t

-

H

t



,

h t is a decision output value of a decision-making model, and H t is an actual output value;

performing gradient optimization on δ, ω i , and b i involved in the second deep neural network by using the error as a gradient direction, performing stepwise optimization on each moment, taking a finally optimized model as the second deep neural network preadapted by the most adapted clustering center, and categorizing information of a current patient in a point cluster of the most adapted clustering center, for subsequent clustering of the first deep neural network.

6 . A processing device of a hematoma aspiration decision-making system for intracerebral hemorrhage, wherein the device comprises:

a processor; and

a memory storing instructions that, when executed by the processor, cause the processing device to perform operations comprising:

acquiring preoperative data for hematoma aspiration from one or more medical sensors or databases, wherein the preoperative data for hematoma aspiration involves a first indicator;

inputting the preoperative data for hematoma aspiration into a first deep neural network to determine an adapted initial hematoma aspiration protocol preoperatively, wherein the first deep neural network is configured to perform multi-center clustering on the preoperative data for hematoma aspiration input into the network and output a hematoma aspiration protocol preadapted by a most adapted clustering center;

acquiring real-time intraoperative data for hematoma aspiration through the hematoma aspiration apparatus, wherein the real-time intraoperative data for hematoma aspiration involves a second indicator, the second indicator comprises an aspiration velocity, a fluid inlet velocity, blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, and a hematoma size;

selecting, from a plurality of pre-trained Long Short-Term Memory (LSTM) networks, a specific LSTM network that is preadapted by and corresponds to the most adapted clustering center determined by the first deep neural network;

inputting the real-time intraoperative data for hematoma aspiration into the a second deep neural network distinct from the first deep neural network, wherein the second deep neural network corresponds to the selected specific LSTM network, so as to determine a decision-making result of the hematoma aspiration protocol and a prediction result of the hematoma aspiration protocol intraoperatively, wherein the second deep neural network is configured to determine, based on the real-time intraoperative data for hematoma aspiration input into the network, a time series control protocol corresponding to a real-time hematoma aspiration process, and the time series control protocol comprises protocol content of the hematoma aspiration protocol in a decision-making aspect and a prediction aspect, wherein the decision-making result comprises control amounts for a fluid inlet amount and a fluid outlet amount by a micro metering pump, and the prediction result comprises a prediction result amount and a prediction decision amount, wherein the prediction result amount comprises blood pressure, blood oxygen, a heart rate, intracranial pressure, a hematoma location, and a hematoma size, and the prediction decision amount comprises the fluid inlet amount and the fluid outlet amount; and

generating, based on the decision-making result, one or more control signals that directly drive the micro metering pump of a hematoma aspiration apparatus to dynamically and physically control a fluid inlet amount through the fluid inlet channel and a fluid outlet amount through the fluid outlet channel during an intracerebral hemorrhage surgery, wherein the micro flowmeter provides instantaneous flow-rate feedback, the gravimeter measures mass change to correct the flow-rate measurement; wherein the method implements a real-time closed-loop control of the hematoma aspiration apparatus by iteratively updating the hematoma aspiration protocol based on newly acquired intraoperative data at each subsequent moment, thereby achieving real-time and high-accuracy control of the hematoma aspiration process;

wherein for the second deep neural network, the time series control protocol is represented as:

{

H

1

,

H

2

,

…

,

H

t

,

…

,

H

T

}

,

 wherein

H t is a control protocol at a moment t, T is a maximum value of t, and T corresponds to a total time,

H

t

=

{

C

t

,

S

t

,

D

t

}

,

 wherein

H t comprises a number of states at the moment t, comprising a control state C t , a static state S t , and a dynamic state D t , specifically:

C t serves to control an action at a next moment, comprising a fluid inlet type, a fluid inlet velocity, and an aspiration velocity,

S t is a descriptor value rather than a specific numerical value, serves as state monitoring, and comprises a heart rate, blood pressure, and blood oxygen,

D t is used to compare with an actual value to correct a control protocol H t+1 at the next moment, comprising intracranial pressure, a hematoma size, and a hematoma location,

the control protocol H t+1 at the next moment is calculated by the following equation:

H

t

+

1

=

δ

[

ω

i

(

H

t

,

x

t

)

+

b

i

]

,

 wherein

X t is an actual state at the moment t, comprising intracranial pressure, a hematoma size, and a hematoma location that are actually measured, δ, ω i , and b i are different network parameters of the second deep neural network, and x t and D t are used to calculate a decision error and correct a decision value at the next moment by using the decision error so as to obtain an actual decision.

7 . A hematoma aspiration decision-making system, comprising a processor and a memory, wherein the memory has a computer program stored therein, and the processor, when calling the computer program in the memory, executes the method according to claim 1 .

8 . A non-transitory computer-readable storage medium, wherein the computer-readable storage medium stores a plurality of instructions; and the instructions are suitable for being loaded by a processor to execute the method according to claim 1 .