# US 2025/0157175 A1 · Personalized Target Selection Method for Non-invasive Neuromodulation Technology

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0157175 A1
- Document ID: US20250157175A1
- Status: Published application
- Kind: A1
- Application number: 19/026,895
- Filed: 2025-01-17
- Published: 2025-05-15
- Earliest claimed: 2023-03-21
- Assignees: The Affiliated Brain Hospital of Nanjing Medical University
- Applicants: The Affiliated Brain Hospital of Nanjing Medical University
- Inventors: Xizhe ZHANG; Fei WANG
- Web page: https://patentgrep.com/patent/US20250157175A1

## CPC

- G06V 10/25: Determination of region of interest [ROI] or a volume of interest [VOI]
- A61B 5/0042: {for the brain}
- A61B 5/055: involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
- A61B 5/7267: {involving training the classification device}
- G06T 7/0014: {using an image reference approach}
- G06T 7/73: using feature-based methods
- G06V 10/26: Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V 10/32: Normalisation of the pattern dimensions
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 20/70: Labelling scene content, e.g. deriving syntactic or semantic representations
- G16H 30/40: for processing medical images, e.g. editing
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- A61B 5/0036: {including treatment, e.g., using an implantable medical device, ablating, ventilating}
- A61B 5/14542: {for measuring blood gases (A61B5/14551 takes precedence)}
- G06T 2207/10088: Magnetic resonance imaging [MRI]
- G06T 2207/20081: Training; Learning
- G06T 2207/30016: Brain
- G06T 2207/30242: Counting objects in image
- G06V 2201/031: of internal organs
- Y02A 90/10: Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- Y02A: TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE

## Abstract

The provided is a personalized target selection method for a non-invasive neuromodulation technology, including: preprocessing functional magnetic resonance imaging (fMRI) data from MRI scan data of a current patient to acquire fMRI brain image feature data; inputting the fMRI brain image feature data into a pre-trained inter-subtype classification model to acquire a subtype label of the current patient and all feature voxels of the subtype label; preprocessing T1-weighted MRI data of structural magnetic resonance imaging (sMRI) data from the MRI scan data of the current patient to acquire a skull outline and a transformation matrix between the sMRI and fMRI data; performing coordinate transformation on the feature voxels, calculating a distance between each voxel on the skull outline and each feature voxel, marking response feature voxels, and counting a number of response feature voxels; and sorting the number of response feature voxels.

## Claims

### Claim 1

A personalized target selection method for a non-invasive neuromodulation technology, comprising:
S10: preprocessing functional magnetic resonance imaging (fMRI) data from MRI scan data of a current patient to acquire preprocessed fMRI brain image feature data;
S20: inputting the preprocessed fMRI brain image feature data into a pre-trained inter-subtype classification model to acquire a subtype label of the current patient and all feature voxels of the subtype label;
S30: preprocessing T1-weighted MRI data of structural magnetic resonance imaging (sMRI) data from the MRI scan data of the current patient to acquire a skull outline; and registering the sMRI and fMRI data to acquire a transformation matrix T; and
S40: performing, based on the transformation matrix T, coordinate transformation on all the feature voxels of the subtype label; calculating a distance between each voxel b on the skull outline and each feature voxel after coordinate transformation; marking a feature voxel with a distance less than a predetermined length as a response feature voxel of b; counting a number of response feature voxels corresponding to each voxel on the skull outline; and selecting a voxel with a largest number of response feature voxels on the skull outline as a candidate target for a corresponding personalized target;
wherein the pre-trained inter-subtype classification model is acquired by training a given classification model based on MRI scan data of a preset number of patients.

### Claim 2 (depends on 1)

The personalized target selection method according to claim 1, wherein before the step S20, the personalized target selection method further comprises:
S001: preprocessing fMRI brain image data of a predetermined number of patients acquired in advance to acquire brain image features from the fMRI data of each patient;
S002: performing dimensionality reduction on each brain image feature to acquire low-dimensional image feature data; clustering, by a clustering algorithm, all the low-dimensional image feature data to acquire k subtypes; and marking each patient with a subtype label; and
S003: taking the low-dimensional image feature data with the subtype label as first training data, and training a classification model to acquire a trained inter-subtype classification model.

### Claim 3 (depends on 2)

The personalized target selection method according to claim 2, wherein
the brain image feature of the fMRI data comprises: amplitude of low frequency fluctuations (ALFF), regional homogeneity (ReHo), or functional connectivity (FC);
the step of performing the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data comprises:
performing, by t-distributed stochastic neighbor embedding (t-SNE), principal component analysis (PCA), or unified manifold approximation and projection (UMAP), the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data; and
the classification model is support vector machine (SVM), or convolutional neural network (CNN).

### Claim 4 (depends on 2)

The personalized target selection method according to claim 2, wherein before the step S30, the personalized target selection method further comprises:
selecting, for a single subtype, brain image feature data of all patients with a corresponding subtype label; acquiring fMRI data of a predetermined number of healthy individuals; acquiring, by a patient data preprocessing method, brain image feature data of the healthy individuals; training a binary classification machine learning (ML) model with a classification label of ill or not; identifying, by a recursive feature elimination (RFE) algorithm, corresponding feature voxels of the subtype; and recording identified feature voxels in SList.

### Claim 5 (depends on 4)

The personalized target selection method according to claim 4, wherein the step S30 comprises:
segmenting, by an MRI brain image skull segmentation algorithm, the T1-weighted MRI data to acquire the skull outline; and
registering, by a brain image analysis method, the sMRI and fMRI data of the current patient to acquire the transformation matrix T;
wherein the fMRI data is partial data of the MRI scan data of the current patient.

### Claim 6 (depends on 5)

The personalized target selection method according to claim 5, wherein the personalized target selection method further comprises:
performing, based on the transformation matrix T, coordinate transformation on each voxel in SList to acquire a feature voxel list SListT of the T1-weighted MRI data in a same coordinate system;
calculating, by a traversal method, a distance between each voxel b on the skull outline and each feature voxel aT in SListT; marking, when the distance between b and aT is less than D, aT as a response voxel of b; and counting a number of response voxels of the voxel b; and
sorting the number of response voxels of the voxel on the skull outline of the patient, and selecting a voxel with a largest number of response feature voxels on the skull outline within D as a candidate therapeutic target.

### Claim 7 (depends on 5)

The personalized target selection method according to claim 5, wherein in the step S10, the fMRI data of the current patient is preprocessed to acquire preprocessed brain image feature data by:
preprocessing the fMRI data of the current patient; selecting a low frequency signal with a frequency between 0.01 Hz and 0.08 Hz; and calculating, by an ALFF calculation method, an ALFF value according to Eq. (1):
ALFF
i
=
∑
i
=
N
1
N
2
⁢
Y
i
N
2
-
N
1
(
1
)
wherein ALFFi denotes the ALFF value of a voxel, comprising a three-dimensional (3D) spatial coordinate point i of the brain image feature data; Yi denotes frequency domain data acquired by performing Fourier transform on time domain data of the voxel i; and N1 and N2 denote data index positions of a discrete frequency spectrum corresponding to selected lowest and highest frequencies, respectively; and
dividing the ALFF value of each voxel by an all-brain mean ALFF value to acquire a standardized ALFF result, and taking the standardized ALFF result as the preprocessed brain image feature data.

### Claim 8 (depends on 1)

A computing device, comprising a memory and a processor, wherein the memory is configured to store a computer program; and the processor is configured to execute the computer program in the memory to implement the steps of the personalized target selection method for the non-invasive neuromodulation technology according to claim 1.

### Claim 9 (depends on 1)

A computer-readable storage medium, configured to store a computer program, wherein the computer program is executed by a processor to implement the steps of the personalized target selection method for the non-invasive neuromodulation technology according to claim 1.

### Claim 10 (depends on 8)

The computing device according to claim 8, wherein before the step S20 of the personalized target selection method, the personalized target selection method further comprises:
S001: preprocessing fMRI brain image data of a predetermined number of patients acquired in advance to acquire brain image features from the fMRI data of each patient;
S002: performing dimensionality reduction on each brain image feature to acquire low-dimensional image feature data; clustering, by a clustering algorithm, all the low-dimensional image feature data to acquire k subtypes; and marking each patient with a subtype label; and
S003: taking the low-dimensional image feature data with the subtype label as first training data, and training a classification model to acquire a trained inter-subtype classification model.

### Claim 11 (depends on 10)

The computing device according to claim 10, wherein in the personalized target selection method, the brain image feature of the fMRI data comprises: amplitude of low frequency fluctuations (ALFF), regional homogeneity (ReHo), or functional connectivity (FC);
the step of performing the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data comprises:
performing, by t-distributed stochastic neighbor embedding (t-SNE), principal component analysis (PCA), or unified manifold approximation and projection (UMAP), the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data; and
the classification model is support vector machine (SVM), or convolutional neural network (CNN).

### Claim 12 (depends on 10)

The computing device according to claim 10, wherein before the step S30 of the personalized target selection method, the personalized target selection method further comprises:
selecting, for a single subtype, brain image feature data of all patients with a corresponding subtype label; acquiring fMRI data of a predetermined number of healthy individuals; acquiring, by a patient data preprocessing method, brain image feature data of the healthy individuals; training a binary classification machine learning (ML) model with a classification label of ill or not; identifying, by a recursive feature elimination (RFE) algorithm, corresponding feature voxels of the subtype; and recording identified feature voxels in SList.

### Claim 13 (depends on 12)

The computing device according to claim 12, wherein the step S30 of the personalized target selection method comprises:
segmenting, by an MRI brain image skull segmentation algorithm, the T1-weighted MRI data to acquire the skull outline; and
registering, by a brain image analysis method, the sMRI and fMRI data of the current patient to acquire the transformation matrix T;
wherein the fMRI data is partial data of the MRI scan data of the current patient.

### Claim 14 (depends on 13)

The computing device according to claim 13, wherein the personalized target selection method further comprises:
performing, based on the transformation matrix T, coordinate transformation on each voxel in SList to acquire a feature voxel list SListT of the T1-weighted MRI data in a same coordinate system;
calculating, by a traversal method, a distance between each voxel b on the skull outline and each feature voxel aT in SListT; marking, when the distance between b and aT is less than D, aT as a response voxel of b; and counting a number of response voxels of the voxel b; and
sorting the number of response voxels of the voxel on the skull outline of the patient, and selecting a voxel with a largest number of response feature voxels on the skull outline within D as a candidate therapeutic target.

### Claim 15 (depends on 13)

The computing device according to claim 13, wherein in the step S10 of the personalized target selection method, the fMRI data of the current patient is preprocessed to acquire preprocessed brain image feature data by:
preprocessing the fMRI data of the current patient; selecting a low frequency signal with a frequency between 0.01 Hz and 0.08 Hz; and calculating, by an ALFF calculation method, an ALFF value according to Eq. (1):
ALFF
i
=
∑
i
=
N
1
N
2
⁢
Y
i
N
2
-
N
1
(
1
)
wherein ALFFi denotes the ALFF value of a voxel, comprising a three-dimensional (3D) spatial coordinate point i of the brain image feature data; Yi denotes frequency domain data acquired by performing Fourier transform on time domain data of the voxel i; and N1 and N2 denote data index positions of a discrete frequency spectrum corresponding to selected lowest and highest frequencies, respectively; and
dividing the ALFF value of each voxel by an all-brain mean ALFF value to acquire a standardized ALFF result, and taking the standardized ALFF result as the preprocessed brain image feature data.

### Claim 16 (depends on 9)

The computer-readable storage medium according to claim 9, wherein before the step S20 of the personalized target selection method, the personalized target selection method further comprises:
S001: preprocessing fMRI brain image data of a predetermined number of patients acquired in advance to acquire brain image features from the fMRI data of each patient;
S002: performing dimensionality reduction on each brain image feature to acquire low-dimensional image feature data; clustering, by a clustering algorithm, all the low-dimensional image feature data to acquire k subtypes; and marking each patient with a subtype label; and
S003: taking the low-dimensional image feature data with the subtype label as first training data, and training a classification model to acquire a trained inter-subtype classification model.

### Claim 17 (depends on 16)

The computer-readable storage medium according to claim 16, wherein in the personalized target selection method, the brain image feature of the fMRI data comprises: amplitude of low frequency fluctuations (ALFF), regional homogeneity (ReHo), or functional connectivity (FC);
the step of performing the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data comprises:
performing, by t-distributed stochastic neighbor embedding (t-SNE), principal component analysis (PCA), or unified manifold approximation and projection (UMAP), the dimensionality reduction on each brain image feature to acquire the low-dimensional image feature data; and
the classification model is support vector machine (SVM), or convolutional neural network (CNN).

### Claim 18 (depends on 16)

The computer-readable storage medium according to claim 16, wherein before the step S30 of the personalized target selection method, the personalized target selection method further comprises:
selecting, for a single subtype, brain image feature data of all patients with a corresponding subtype label; acquiring fMRI data of a predetermined number of healthy individuals; acquiring, by a patient data preprocessing method, brain image feature data of the healthy individuals; training a binary classification machine learning (ML) model with a classification label of ill or not; identifying, by a recursive feature elimination (RFE) algorithm, corresponding feature voxels of the subtype; and recording identified feature voxels in SList.

### Claim 19 (depends on 18)

The computer-readable storage medium according to claim 18, wherein the step S30 of the personalized target selection method comprises:
segmenting, by an MRI brain image skull segmentation algorithm, the T1-weighted MRI data to acquire the skull outline; and
registering, by a brain image analysis method, the sMRI and fMRI data of the current patient to acquire the transformation matrix T;
wherein the fMRI data is partial data of the MRI scan data of the current patient.

### Claim 20 (depends on 19)

The computer-readable storage medium according to claim 19, wherein the personalized target selection method further comprises:
performing, based on the transformation matrix T, coordinate transformation on each voxel in SList to acquire a feature voxel list SListT of the T1-weighted MRI data in a same coordinate system;
calculating, by a traversal method, a distance between each voxel b on the skull outline and each feature voxel aT in SListT; marking, when the distance between b and aT is less than D, aT as a response voxel of b; and counting a number of response voxels of the voxel b; and
sorting the number of response voxels of the voxel on the skull outline of the patient, and selecting a voxel with a largest number of response feature voxels on the skull outline within D as a candidate therapeutic target.
