CN113180650B

CN113180650B - Near-infrared brain imaging atlas identification method

Info

Publication number: CN113180650B
Application number: CN202110099644.0A
Authority: CN
Inventors: 刘破资; 戎戈; 刘思中
Original assignee: Beijing Yingguo Xingkong Technology Co ltd
Current assignee: Beijing Yuguo Technology Co.,Ltd.
Priority date: 2021-01-25
Filing date: 2021-01-25
Publication date: 2023-02-07
Anticipated expiration: 2041-01-25
Also published as: CN113180650A

Abstract

The present disclosure discloses a near-infrared brain imaging atlas analysis method, which includes: extracting single-channel full-time-sequence map data of relative concentrations of deoxyhemoglobin and oxyhemoglobin of a plurality of channels of a forehead calibration brain area of a patient under the stimulation of a cognitive task, and constructing a detection curve of the deoxyhemoglobin and the oxyhemoglobin according to the map data; sampling the detection curve, taking the total detection time T as a horizontal axis to perform subarea sampling, and performing normalization processing on a vertical axis to obtain a final data set; and constructing a neural network algorithm model according to the standard map curve, respectively inputting the final data sets of the channels into the neural network algorithm model, judging the proportion of the standard map in the detection curves corresponding to the channels, and obtaining a judgment result according to the preset corresponding relation between the preset proportion and the standard maps. The near-infrared brain imaging atlas analysis method provided by the invention can provide reference for clinical diagnosis.

Description

Near-infrared brain imaging atlas identification method

Technical Field

The invention relates to an image identification method, in particular to a near-infrared brain imaging atlas identification method.

Background

Near-infrared spectroscopy (NIRS) utilizes absorption and scattering relationships between two Near-infrared lights and deoxyhemoglobin (a.k.a.Deoxy-Hb) and oxyhemoglobin (Oxy-Hb) in brain tissues, so that researchers can quantitatively observe relative concentration changes of Oxy-Hb and Deoxy-Hb in cortical tissues in a non-invasive mode, and brain activity is reflected indirectly. In the 80's of the last century, NIRS gradually entered the medical field due to the development of optical fiber technology. Farrari and colleagues reported in 1985 the first study of quantitative measurement of cerebral blood oxygen in humans using near infrared spectroscopy. Researchers have since begun to use functional near infrared spectroscopy (fNIRS) for relevant medical discussions, and clinical studies have been in progress for nearly 30 years.

NIRS has been widely used in the study of cerebral cortical dysfunction in psychiatric disorders for the last decade. Compared with other nerve function images such as PET, SPECT, fMRI and other technologies, NIRS is limited to observe functional activity of the cerebral cortex about 2cm below the probe, but has the advantages of being relatively light and concise, easily meeting measurement conditions and being high in time resolution, and allows a subject to be tested in a relatively comfortable environment, so that the subject can keep still or perform tasks under relatively natural conditions, and therefore relatively real data can be obtained.

In recent years, functional imaging studies have confirmed abnormal frontal and temporal lobe blood perfusion in patients with depression. The Functional magnetic resonance imaging (fMRI) literature on depression suggests that blood flow in the left frontal lobe, temporal lobe, amygdala cingulate gyrus, etc. of depressed patients is reduced, and the original hypometabolic rate and hypoperfusion areas are improved after stimulation and brain function activation. A large number of PET and SPECT researches find that the blood flow of the prefrontal cortex of a depression patient in the acute stage is obviously reduced, so that the neuron malnutrition is caused, and the depression symptom is obvious.

The NIRS is used by foreign scholars to verify the cerebral cortex dysfunction of depression patients, and the results show that the specific range of cerebral cortex activation reduction is different due to different severity and research methods of selected cases, and although no consistent result exists, most research results show that the prefrontal lobe and adjacent temporal lobe are low in function. This phenomenon points to a number of clinical studies that are performed later.

The Verbal Fluency Task (VFT) is a stimulating Task that measures the brain activity of the subject's prefrontal lobes by producing the number of words in a valid time and in a prescribed category. It mainly relates to the speed of extraction of memory information and attention energy. The literature reports that the fluency of speech is closely related to a central execution system, and partial research supports that VFT is mainly related to frontal lobe function. VFT is widely used in studies to assess cognitive function deficits in patients with mental disorders, and in 2008, the japanese scholari Hori et al considered that rhyme-related VFT is most effective in NIRS to activate the prefrontal lobe for a range of tasks that are useful for mental disorders.

More basic researches show that in a normal state, the activity of neurons in a brain region increases the oxygen consumption of the region, and meanwhile, cerebral blood volume and blood flow carry a large amount of oxyhemoglobin, so that the concentration of local oxyhemoglobin is obviously increased, the concentration of deoxyhemoglobin is fluctuated along with the concentration of oxyhemoglobin, the concentration of total hemoglobin is increased, the oxygen saturation is increased, and the brain region is in a high-oxygen-concentration blood perfusion state. In pathological conditions, the activity of neurons in a brain region increases the oxygen consumption of the region, the blood volume and the blood flow of the brain are not obviously or slightly reduced, a large amount of oxyhemoglobin is not supplied, the concentration of local deoxyhemoglobin is increased, the concentration of oxyhemoglobin fluctuates or is reduced, the concentration of total hemoglobin is reduced, the oxygen saturation of blood is reduced, and the brain region is in a low-oxygen-concentration blood flow perfusion state. These two states can be effectively reproduced in the prefrontal lobe of the cerebral cortex by cognitive stimulation tasks such as VFT.

In recent years, there were domestic scholars who monitored the change in hemoglobin concentration in prefrontal cortex (PFC) under speech fluency task (VFT) in 30 MDD patients with anxiety, obsessive-compulsive symptoms using NIRS technology, and then analyzed the NIRS data in PFC in relation to the severity of depression, anxiety and obsessive-compulsive symptoms. The symptoms of depression, anxiety and obsessive-compulsive disorder are assessed using the Hamilton Depression Scale (HAMD), the Hamilton anxiety Scale (HAMA) and the Yale-Brown obsessive-compulsive Scale (Y-BOCS), respectively. This study confirmed the presence of low activation in the lateral prefrontal cortex, lower prefrontal cortex, of MDD patients. Later team studies found that changes in oxygenated hemoglobin concentration (oxy-Hb) in the right prefrontal cortex correlated with the severity of anxiety symptoms. At the same time, no statistically relevant results were found with respect to the severity of obsessive-compulsive symptoms. The results of the study also show that MDD patients with anxiety and obsessive-compulsive symptoms exhibit a state of low activation of the prefrontal cortex in NIRS. In addition, the right prefrontal cortex is associated with anxiety symptoms, and the bilateral prefrontal cortex and the anterolateral prefrontal cortex are associated with depression symptom severity. But unlike depression, anxiety, obsessive compulsive symptoms may have different biological properties in the prefrontal cortex function. In addition, hypoactivation of hemoglobin, which is associated with depression, is found in blepharospasm patients, menopausal depressed patients, somatoform pain disorder patients, and chronic insomnia patients.

In subsequent researches, a domestic research team carries out quantitative investigation on the correlation between the fNIRS blood oxygen activation pattern and the depressive anxiety obsessive-compulsive symptom under multiple cognitive tasks in 2012 to 2015, the team carries out classification analysis on a brain imaging map of a typical case from the aspect of symptomatology, compares the phase, waveform, amplitude, frequency and trend characteristics of a hyperbolic task state of 45 channels Oxy-Hb and Deoxy-Hb in a brain area calibrated by forehead when a normal control group and a depressive disorder group and an anxiety disorder group carry out multiple equivalent language fluency tasks, and carries out preliminary analysis. Fig. 1 shows some typical known near-infrared brain imaging maps. Wherein:

fig. 1-a, normal-phase hyperbolic parallel waveform (normal-phase waveform): in the positive phase, oxy-Hb is on the upper part, deoxy-Hb is on the lower part, the amplitude of a hyperbolic curve floats about 30% of an observation window, and the waveform is regular and is similar to a sine wave. Normal people have normal energy and physical strength, normal working state of life and study and harmonious interpersonal relationship. No physical symptoms: no insomnia, dizziness, fatigue or headache in nearly 2 weeks.

Fig. 1-b, normal phase high amplitude waveform (generalized anxiety waveform): in the positive phase, oxy-Hb is above, deoxy-Hb is below, the amplitude of a hyperbolic curve is obviously higher than 30% of an observation window, and fluctuation has a falling trend after a single task state is finished. Psychological symptoms: the people always worry about what unfortunate things happen, such as rare tension, fear, uneasiness, no treading, uneasy sitting, eager, unsmooth sitting and easy spleen qi generation. Is more sensitive to the environment. Somatic symptoms: often with body symptoms such as paroxysmal headache, palpitation, and muscle jumping.

FIG. 1-c, normal phase open-like waveform (anxiety-obsessive-compulsive waveform): the positive phase, oxy-Hb is on the upper part, deoxy-Hb is on the lower part, the fluctuation has no falling trend after the single task state is finished, and the hyperbola is in a positive phase opening shape. Psychological symptoms: people who want to see what repeatedly want the life events related to the people, and the life events are sensitive to self feeling and environment, uneasy in mind, tense and easy to lose control of emotion and irritate. Somatic symptoms: sometimes accompanied by migratory pain, numbness in the limbs, pain in the muscles and joints, and the like.

FIG. 1-d, negative phase hyperbolic parallel waveform (depressive hysteresis waveform): deoxy-Hb is above, oxy-Hb is below or has no obvious positive and negative phase characteristics, and the amplitude of the hyperbola is low and flat below 30% of an observation window. Psychological symptoms: the brain reaction is slow and dull, the brain is unclear and uninterested, the brain is oppressed and unwilling to speak, the speaking sound is low, and the memory is reduced. Somatic symptoms: dizziness, heaviness of the head, insomnia, easy fatigue and the like.

Figure 1-e, negative phase hyperbolic intersecting waveform (depression anxiety waveform): deoxy-Hb is above, oxy-Hb is below or has no obvious positive and negative phase characteristics, the amplitude of the hyperbola is above 30% of the observation window or the drift is unstable. Psychological symptoms: unstable emotion, distraction, distress, hopelessness, worthless, and easy crying and paradoxical thinking. The attention is not concentrated, the mind is disordered and the mind is not resistant. Somatic symptoms: general malaise, loss of appetite and poor appetite.

FIG. 1-f, negative phase hyperbolic open-ended waveform (Depression forcing waveform): negative phase, deoxy-Hb is above, oxy-Hb is below, hyperbolas have superposition, and fluctuation does not fall back after a single task state is finished, so that the negative phase is open. Psychological symptoms: the unpleasant things are recalled repeatedly, and the feeling of depression, joy, no use and depression are accompanied. The grief over what has happened. Somatic symptoms: there are cases where somatization symptoms such as chest discomfort and mouth discomfort occur.

Therefore, on the basis of a large number of researches, the standard map obtained by the domestic research team has a relatively clear practical significance on the auxiliary diagnosis of mental diseases such as extensive anxiety, anxiety and obsessive-compulsive disorder, depression retardation, depression anxiety, depression and depression compulsion. In fact, the current research results only can help related medical workers to establish cognition according to typical maps of various mental diseases, and the diagnosis efficiency cannot be effectively improved or the misdiagnosis probability cannot be effectively reduced. In the actual diagnosis process, the influence of subjective factors is still difficult to avoid for patients and medical doctors. In other words, currently, when a diagnostician makes a diagnosis according to a near-infrared brain imaging atlas of an actual patient, the subjective judgment is still made based on experience. In addition, in the conventional diagnosis and treatment process, when the infrared brain imaging atlas of the patient is diagnosed, another defect of manual judgment except the influence of subjective consciousness is that the diagnosis efficiency is low, and the diagnosis result may be delayed to a large extent for a large-scale hospital.

It can be seen that there is a need for improvement in the above-mentioned problems of the prior art.

Disclosure of Invention

In view of the above problems in the prior art, the present invention is directed to a near-infrared brain imaging atlas analysis method for assisting a psychiatrist in diagnosis of mental diseases, which can significantly reduce the subjective influence of the medical doctor.

In order to achieve the above object, an embodiment of the present invention provides a near-infrared brain imaging atlas analysis method, including:

extracting single-channel full-time-sequence map data of relative concentrations of deoxyhemoglobin and oxyhemoglobin of a plurality of channels of a forehead calibration brain area of a patient under the stimulation of a cognitive task, and constructing a detection curve of the deoxyhemoglobin and the oxyhemoglobin according to the map data;

sampling the detection curve, taking the total detection time T as a horizontal axis to perform subarea sampling, and performing normalization processing on a vertical axis to obtain a final data set;

and constructing a neural network algorithm model according to the standard map curve, respectively inputting the final data sets of the plurality of channels into the neural network algorithm model, judging the proportion of the standard map in the detection curves corresponding to the plurality of channels, and obtaining a judgment result according to the preset corresponding relation between the preset proportion and the plurality of standard maps.

Preferably, constructing a detection curve of deoxyhemoglobin and oxyhemoglobin from the atlas data comprises:

gaussian low-pass filtering is carried out on the detection curve;

setting an observation interval on the longitudinal axis of the detection curve according to a near-infrared brain imaging spectrum detection instrument;

the normalization process is performed in the observation interval.

Preferably, a detection curve of the deoxyhemoglobin and the oxyhemoglobin is constructed according to the atlas data, and the method further comprises removing a drift signal exceeding a first preset value of the observation interval and reconstructing the detection curve;

wherein sampling the detection curve comprises sampling the reconstructed detection curve.

Preferably, the divisional sampling of the total detection time T on the horizontal axis includes sampling the respective proteins deoxyhemoglobin and oxyhemoglobin in the detection curve, and the divisional sampling of the total detection time T on the horizontal axis corresponds to 1/6,1/3,1/2,2/3, and 5/6 of the horizontal axis.

Preferably, when the normalization process is performed, the final data set is determined by:

D _g ＝{D _oxy ,D _dxy ,D _p }

wherein D is _oxy For oxyhemoglobin sampling results, D _dxy For deoxygenated hemoglobin sampling results, R _o To observe the interval, D _p The peripheral data includes the demographic data (such as age and sex) generated by detection.

Preferably, the oxygenated hemoglobin sampling result is determined by:

the deoxygenated hemoglobin sampling result is determined by:

wherein

To is that

And

to is that

Respectively corresponding to different subarea samples.

Preferably, the constructing of the neural network algorithm model comprises: and using six forward neural networks, wherein each forward neural network respectively has six different standard maps, and then carrying out supervised training on each forward neural network until the error variance is less than a second preset value.

Preferably, the method for constructing the neural network algorithm model further comprises constructing an expert system, wherein the output N of each forward neural network after training is used as the input of the expert system, and the value range of each output N is {0,1}.

Preferably, the expert system is constructed, and N with the highest output value is selected preferentially _max And removing all N less than 0.5; the data between 0.5,1.0 is output as candidate data.

Preferably, the proportion of the standard map in the detection curves corresponding to the plurality of channels is determined as follows:

Pk＝{C _Nk /C-Ci,k∈1,2,3,4,5,6}

wherein Pk is the specific gravity, C is the total number of channels, CNk is the number of channels with different spectrum types, and Ci is the number of channels without spectrum types.

Compared with the prior art, the near-infrared brain imaging atlas analysis method provided by the invention can be used for analyzing and identifying the multichannel near-infrared brain imaging atlas of the patient by means of a neural network algorithm and an expert system, and provides reference for clinical diagnosis. Compared with pure manual identification, the influence of subjective consciousness of a doctor is reduced, and compared with a general intelligent algorithm, the result is more reliable.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

This document provides an overview of various implementations or examples of the technology described in this disclosure, and is not a comprehensive disclosure of the full scope or all features of the disclosed technology.

Drawings

FIG. 1 is a standard pattern (where FIG. 1-a is normal pattern, FIG. 1-b is generalized anxiety, FIG. 1-c is anxious obsessive compulsive, FIG. 1-d is depressive lag, FIG. 1-e is depressive anxiety and FIG. 1-f is depressive obsessive compulsive).

Fig. 2 is a general flowchart of one embodiment of the near-infrared brain imaging atlas analysis method of the invention.

Fig. 3 is a model structure diagram of a single forward neural network in the near-infrared brain imaging atlas analysis method of the present invention.

Fig. 4 is a flowchart of a complete implementation of the near-infrared brain imaging atlas analysis method of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions of the embodiments of the present disclosure will be described below clearly and completely with reference to the accompanying drawings of the embodiments of the present disclosure.

It is to be understood that the described embodiments are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the disclosure without any inventive step, are within the scope of protection of the disclosure.

To maintain the following description of the embodiments of the present disclosure clear and concise, a detailed description of known functions and known components have been omitted from the present disclosure.

As shown in fig. 2, a near-infrared brain imaging atlas analysis method provided in an aspect of an embodiment of the present invention includes:

in this step, the cognitive Task is stimulated in addition to the verbosity Task (VFT), which is a stimulating Task that measures the subject's prefrontal lobe brain activity by producing the number of words in a prescribed category and within an effective time. Then, four groups of neutral words (fruits, vegetables, household appliances and animals with four legs) are selected from the common sense of the Wechsler intelligence scale, so that the patient speaks the name of a specific object after the words are associated, the description time is 30 seconds, the rest time is 30 seconds, and the four groups are 4 minutes in total. Four groups of neutral words in the task, the inductive ability to understand the subject's linguistic concept has no emotional color, and is used for basic symptom map analysis.

In addition, when the method is executed, the cognitive tasks can also comprise cognitive subtasks, rest is carried out between the subtasks at intervals of 30 seconds, and the sequence is ordered according to the psychological load of the tasks and the requirements of experimental needs and clinical psychological significance. The method specifically comprises the following steps: a rest state: the pulse rate of the testee in the resting state is recorded, and the pulse rate of the testee is recorded. The test was followed by resting for 5 minutes and recording the resting for 30 seconds. (II) Emotion Picture Task (EPT): selecting 15 emotional pictures from the international emotional picture task atlas, wherein 5 positive pictures, 5 neutral pictures and 5 negative pictures are displayed in each picture for 4 seconds, and the key evaluation is pleasant, neutral and unpleasant, and the time is 2 seconds, and is 1 minute and 30 seconds in total. The method is used as an environment adaptability task of a cognitive task, and pre-experimental application is generated for the following language smoothness task, so that a detected person has an adaptability period, and the stability is ensured; the task has low stimulation intensity, and can relax the subject and reduce the tension feeling before the task of fluent language. (III) finger motion task (FTT): the patient finishes the finger-to-finger tasks of the thumb and the other four fingers according to the requirements of the screen prompt, the patient moves repeatedly from the index finger to the little finger and then from the little finger to the index finger in sequence for 30 seconds, the patient rests for 30 seconds, and the left hand and the right hand are alternately performed once. Total 2 groups, 2 minutes. And (4) adaptability task. (IV) symptom stimulation task (SPT): some pictures (picture contents are according to experimental psychology principles, including stress rescue scenes, headache symptom pictures, chest pain and family violent cartoons) appear on the screen, and the tested person needs to describe the seen picture contents and feelings by simple words or 3-5 phrases, wherein the description time is 30 seconds, the rest period is 30 seconds, each group is 60 seconds, 4 groups are provided, and the total time is 4 minutes.

Furthermore, after detecting 45 channels of the brain region calibrated on the forehead of the patient to obtain full time-series chart spectrum data, further processing can be performed for obtaining a detection curve, specifically: performing Gaussian low-pass filtering on the detection curve; setting an observation interval on the longitudinal axis of the detection curve according to a near-infrared brain imaging spectrum detection instrument; the normalization process is performed in the observation interval. Constructing a detection curve of the deoxyhemoglobin and the oxyhemoglobin according to the atlas data, removing a drift signal exceeding a first preset value of the observation interval, and reconstructing the detection curve; wherein sampling the detection curve comprises sampling the reconstructed detection curve.

Then, in the method of the present invention, sampling may be performed on the detection curve in the previous step, performing divisional sampling with the total detection time T as a horizontal axis, and performing normalization processing on a vertical axis to obtain a final data set; specifically, the divisional sampling of the total detection time T on the horizontal axis includes sampling the respective proteins deoxyhemoglobin and oxyhemoglobin in the detection curve, and the divisional sampling of the total detection time T on the horizontal axis corresponds to 1/6,1/3,1/2,2/3, and 5/6 of the horizontal axis. While the normalization process is performed, the final data set is determined as follows:

D _g ＝{D _oxy ,D _dxy ,D _p }

wherein D is _oxy For oxyhemoglobin sampling results, D _dxy As a result of deoxygenated hemoglobin sampling, R _o To observe the interval, D _p The peripheral data includes the demographic data (such as age and sex) generated by detection. The oxygenated hemoglobin sampling result is determined by:

the deoxygenated hemoglobin sampling result is determined by:

wherein

To is that

And

to is that

Respectively corresponding to different subarea samples.

Furthermore, in the invention, a neural network algorithm model can be constructed according to a standard map curve, the final data sets of a plurality of channels are respectively input into the neural network algorithm model, the proportion of the standard map in the detection curves corresponding to the channels is judged, and a judgment result is obtained according to the preset corresponding relation between the preset proportion and the standard maps. The proportion of the standard map in the detection curves corresponding to the channels is determined in the following mode:

Pk＝{C _Nk /C-Ci,k∈1,2,3,4,5,6}

wherein Pk is specific gravity, C is total channel number, CNk is channel number of different spectrum types, and Ci is channel number of no spectrum type.

The method of the present invention is further illustrated by the complete flow shown in FIG. 4 as follows:

the complete implementation process comprises the following steps:

1. map (data) preprocessing: the method firstly decomposes single-channel full-task time sequence map data into Hb ^oxy Curve (relative change in oxygenated hemoglobin concentration) and Hb ^dxy Curve (relative change in deoxyhemoglobin concentration), gaussian low-pass filtering at a frequency of 5Hz and setting a profile observation interval R according to the instrument used _o (one each of positive and negative)Interval) adjustment map. Finally, any adjacent point is exceeded

Removing the drift signal of (1), reconstructing Hb ^oxy And Hb ^dxy Curve line.

2. Map (data) feature extraction: the algorithm then samples the reconstructed dual blood oxygen curves, samples the two protein curves at 1/6,1/3,1/2,2/3,5/6 of the total time T (horizontal axis), and samples the two protein curves at the observation interval R _o Normalized within (vertical axis) to produce a final data set D _g ：

D _g ＝{D _oxy ,D _dxy ,D _p }

Wherein D is _oxy For oxyhemoglobin sampling results, D _dxy For deoxygenated hemoglobin sampling results, D _p The peripheral data include the demographic data (age and sex) generated by the test.

3. Training a neural network: the present algorithm uses six forward neural networks (FFNs), each FFN identifying one of the spectral patterns shown in fig. 1, as shown in fig. 3. The number of input nerves per net is thus 14 and the number of output nerves is 1 (fig. 3). Each network will be trained using Back Propagation (BP) supervised training, respectively using 20-50 different demographic and semi-control maps until the error variance meets the requirement of less than 1%.

4. Constructing an expert system: the discrimination system contains six of the aforementioned FFNs, and the output N of each FFN is taken as input to the expert system. Each N has a value range of 0, 1. The system preferentially selects N with the highest output value _max And rejecting all N less than 0.5, with outputs between 0.5,1.0 being candidates, and the final output of the system (i.e. the final atlas classification for a single channel) being derived from the following table (only some of the ideal conditions listed):

N1

N2

N3

N4

N5

N6

output of

Means of

Max

Rejecting

Removing

Rejecting

Removing

1

Spectrum type 1

Rejecting

Max

Removing

Rejecting

2

Pattern 2

Removing

Max

Rejecting

Removing

3

Spectrum type 3

Removing

Max

Removing

4

Pattern 4

Removing

Max

Removing

5

Spectrum type 5

Removing

Max

6

Spectrum type 6

Removing

0

Non-music score type

And finally, according to the total channel number C (spectrum type number CN, channel number Ci without spectrum type) of the spectrum, calculating the specific gravity Pk of each spectrum type in all channels, wherein Pk is a final reference value for outputting a judgment result. For example, if the total number of judged results with an output of 1 is 25, P1=50%, such as C =50, ci = 0. The judgment result tends to the spectrum type 1, and by analogy, the patient spectrum type can be quickly judged according to the Pk values respectively, so as to assist the diagnosis of doctors.

The above embodiments are only exemplary embodiments of the present invention, and are not intended to limit the present invention, and the scope of the present invention is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present invention, and such modifications and equivalents should also be considered as falling within the scope of the present invention.

Claims

1. The near-infrared brain imaging atlas analysis method includes:

establishing a neural network algorithm model according to a standard map curve, respectively inputting the final data sets of the channels into the neural network algorithm model, judging the proportion of the standard map in the detection curves corresponding to the channels, and obtaining a spectrum type judgment result according to the preset corresponding relation between the preset proportion and the standard maps;

wherein constructing a detection curve of deoxyhemoglobin and oxyhemoglobin from the atlas data comprises:

performing Gaussian low-pass filtering on the detection curve;

the normalization processing is carried out in the observation interval;

constructing a detection curve of the deoxyhemoglobin and the oxyhemoglobin according to the atlas data, and removing a drift signal exceeding a first preset value of the observation interval to reconstruct the detection curve;

wherein sampling the detection curve comprises sampling the reconstructed detection curve;

taking the total detection time T as a horizontal axis to perform partition sampling, wherein the partition sampling comprises respectively sampling detection curves of deoxyhemoglobin and oxyhemoglobin, and when the total detection time T is taken as the horizontal axis to perform partition, the total detection time T respectively corresponds to 1/6,1/3,1/2,2/3 and 5/6 of the horizontal axis;

when the normalization process is performed, the final data set is determined by:

wherein D is _oxy For oxyhemoglobin sampling results, D _dxy For deoxygenated hemoglobin sampling results, R _o To observe the interval, D _p The peripheral data comprises the demographic data generated by detection;

the oxygenated hemoglobin sampling result is determined by:

；

the deoxygenated hemoglobin sampling result is determined by:

；

wherein

To

，

To

Respectively corresponding to different subarea sampling;

determining the proportion of the standard map in the detection curves corresponding to the channels by the following method:

wherein Pk is specific gravity for outputting the spectrum type determination result, C is total channel number, C _Nk The number of channels with different spectrum types, and Ci is the number of channels without spectrum types;

constructing a neural network algorithm model, comprising: using six forward neural networks, wherein each forward neural network corresponds to six different standard maps respectively, and then performing supervised training on each forward neural network until the error variance is smaller than a second preset value;

constructing a neural network algorithm model, and constructing an expert system, wherein the output N of each trained forward neural network is used as the input of the expert system, and the value range of each output N is {0,1};

constructing an expert system, and preferably selecting N with the highest output value _max And removing all N less than 0.5; the data between 0.5,1.0 is output as candidate data.