Cerebral perfusion state classification apparatus and method, device, and storage medium

1. A cerebral perfusion state classification apparatus, comprising:

a transceiving module, configured to acquire cervical blood flow data from an ultrasound data collecting device; and

a processor, configured to determine cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data; and

also configured to classify cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data, wherein classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data comprises extract blood perfusion characteristics of a plurality of brain regions from the cerebral perfusion data; and

determine a cerebral perfusion state type to which each of the plurality of brain regions belongs based on the blood perfusion characteristics and blood perfusion characteristic thresholds corresponding to cerebral perfusion state types; and

wherein when determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data, the processor is specifically configured to:

extract cervical blood flow characteristics from the cervical blood flow data; and

input the cervical blood flow characteristics into a pre-trained network model to obtain cerebral perfusion data corresponding to the cervical blood flow characteristics,

wherein the network model is trained based on cervical blood flow characteristic samples and cerebral perfusion data samples;

wherein the network model comprises: a Seq2Seq model comprising an encoder and a decoder constructed based on long short-term memory (LSTM);

the network model further comprises: a forget gate (f t ), an input gate (i t ), and an output gate (o t );

the forget gate (f t ) is realized as: f t =σ(w f ·h t-1 +u f ·x t +b f ); x t is a current input, h t-1 is a previous output,

the input gate (i t ) is realized as: i t =σ(w i ·h t-1 +u i ·x t +b i ), {tilde over (c)} t =tan h(w c ·h t-1 +u c ·x t +b c ), and c t =f t ⊙C t +i t ⊙{tilde over (C)} t ; C t is a current cell state, {tilde over (C)} t is a new information; and

the output gate o t is realized as: o t =σ(w o ·h t-1 +u o ·x t +b o ), and h t =o t tan h⊙(C t ).

2. The apparatus according to claim 1 , wherein the processor is further configured to:

receive a brain magnetic resonance image obtained by arterial spin labeling;

divide the brain magnetic resonance image into a plurality of brain regions; and

take cerebral perfusion data of each of the plurality of brain regions as the cerebral perfusion data samples.

3. The apparatus according to claim 1 , wherein the network model comprises: a Seq2Seq model comprising an encoder and a decoder;

and the Seq2Seq model comprises: an encoder and a decoder that are constructed based on long short-term memory (LSTM).

4. The apparatus according to claim 1 , wherein the blood perfusion characteristics comprise cerebral blood flow; and

the processor is further configured to: set cerebral blood flow thresholds corresponding to the cerebral perfusion state types.

5. The apparatus according to claim 1 , wherein the cervical blood flow data comprise one or a combination of: cervical vessel blood flow data and vascular lumen morphology change data.

6. A cerebral perfusion state classification method, comprising:

acquiring cervical blood flow data from an ultrasound data collecting device;

determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data; and

classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data, wherein classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data comprises extracting blood perfusion characteristics of a plurality of brain regions from the cerebral perfusion data; and

determining a cerebral perfusion state type to which each of the plurality of brain regions belongs based on the blood perfusion characteristics and blood perfusion characteristic thresholds corresponding to cerebral perfusion state types; and

wherein when determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data, the method further comprises:

extracting cervical blood flow characteristics from the cervical blood flow data; and

inputting the cervical blood flow characteristics into a pre-trained network model to obtain cerebral perfusion data corresponding to the cervical blood flow characteristics,

wherein the network model is trained based on cervical blood flow characteristic samples and cerebral perfusion data samples;

wherein the network model comprises: a Seq2Seq model comprising an encoder and a decoder constructed based on long short-term memory (LSTM);

the network model further comprises: a forget gate (f t ), an input gate (i t ), and an output gate (o t );

the forget gate (f t ) is realized as: f t =σ(w f ·h t-1 +u f ·x t +b f ); x t is a current input, h t-1 is a previous output,

the input gate (i t ) is realized as: i t =σ(w i ·h t-1 +u i ·x t +b i ), {tilde over (c)} t =tan h(w c ·h t-1 +u c ·x t +b c ), and c t =f t ⊙C t +i t ⊙{tilde over (C)} t ; C t is a current cell state, {tilde over (C)} t is a new information; and

the output gate o t is realized as: o t =σ(w o ·h t-1 +u o ·x t +b o ), and h t =o t tan h⊙(C t ).

7. An electronic device, comprising: a memory and a processor, wherein,

the memory is configured to store a program; and

the processor is coupled with the memory and configured to execute the program stored in the memory so as to:

acquire cervical blood flow data from an ultrasound data collecting device;

determine cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data; and

classify cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data, wherein classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data comprises extract blood perfusion characteristics of a plurality of brain regions from the cerebral perfusion data; and

determine a cerebral perfusion state type to which each of the plurality of brain regions belongs based on the blood perfusion characteristics and blood perfusion characteristic thresholds corresponding to cerebral perfusion state types; and

wherein when determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data, the processor is specifically configured to:

extract cervical blood flow characteristics from the cervical blood flow data; and

input the cervical blood flow characteristics into a pre-trained network model to obtain cerebral perfusion data corresponding to the cervical blood flow characteristics, wherein the network model is trained based on cervical blood flow characteristic samples and cerebral perfusion data samples;

wherein the network model comprises: a Seq2Seq model comprising an encoder and a decoder constructed based on long short-term memory (LSTM);

the network model further comprises: a forget gate (f t ), an input gate (i t ), and an output gate (o t );

the forget gate (f t ) is realized as: f t =σ(w f ·h t-1 +u f ·x t +b f ); x t is a current input, h t-1 is a previous output,

the input gate (i t ) is realized as: i t =σ(w i ·h t-1 +u i ·x t +b i ), {tilde over (c)} t =tan h(w c ·h t-1 +u c ·x t +b c ), and c t =f t ⊙C t +i t ⊙{tilde over (C)} t ; C t is a current cell state, {tilde over (C)} t is a new information; and

the output gate o t is realized as: o t =σ(w o ·h t-1 +u o ·x t +b o ), and h t =o t tan h⊙(C t ).

8. A non-transitory computer storage medium, configured to store a computer program that, when executed on a computer, performs the following method:

acquiring cervical blood flow data from an ultrasound data collecting device;

determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data; and

classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data, wherein classifying cerebral perfusion states of a plurality of brain regions based on blood perfusion characteristics of the plurality of brain regions in the cerebral perfusion data comprises extracting blood perfusion characteristics of a plurality of brain regions from the cerebral perfusion data; and

determining a cerebral perfusion state type to which each of the plurality of brain regions belongs based on the blood perfusion characteristics and blood perfusion characteristic thresholds corresponding to cerebral perfusion state types; and

wherein when determining cerebral perfusion data corresponding to the cervical blood flow data based on the cervical blood flow data and a mapping relationship between the cervical blood flow data and the cerebral perfusion data, the method further comprises:

extracting cervical blood flow characteristics from the cervical blood flow data; and

inputting the cervical blood flow characteristics into a pre-trained network model to obtain cerebral perfusion data corresponding to the cervical blood flow characteristics,

wherein the network model is trained based on cervical blood flow characteristic samples and cerebral perfusion data samples;

wherein the network model comprises: a Seq2Seq model comprising an encoder and a decoder constructed based on long short-term memory (LSTM);

the network model further comprises: a forget gate (f t ), an input gate (i t ), and an output gate (o t );

the forget gate (f t ) is realized as: f t =σ(w f ·h t-1 +u f ·x t +b f ); x t is a current input, h t-1 is a previous output,

the input gate (i t ) is realized as: i t =σ(w i ·h t-1 +u i ·x t +b i ), {tilde over (c)} t =tan h(w c ·h t-1 +u c ·x t +b c ), and c t =f t ⊙C t +i t ⊙{tilde over (C)} t ; C t is a current cell state, {tilde over (C)} t is a new information; and

the output gate o t is realized as: o t =σ(w o ·h t-1 +u o ·x t +b o ), and h t =o t tan h⊙(C t ).