Brain imaging processing system for treatment-resistant depression and method thereof

Abstract

A brain imaging system for treatment-resistant depression includes a communication module, an image processing module, a machine learning module, and a feature analysis module. The communication module is connected to a hospital host computer. The image processing module converts medical images into a new format and aligns them to a standard brain template for spatial normalization. It then uses a standardized uptake value to correct the normalized medical images, generating a preprocessed medical image. This preprocessed image is then divided into multiple brain regions using an automatic anatomical annotation template, and standardized using the average metabolic value of each brain region as a benchmark, generating a standardized preprocessed medical image. The machine learning module uses multiple machine learning models to analyze the standardized preprocessed medical image for treatment-resistant depression based on a first medical feature, obtaining a diagnosis of treatment-resistant depression. The feature analysis module performs an influence analysis on each brain region in the diagnosis of treatment-resistant depression, analyzing the individual influence of each brain region in each machine learning model on treatment-resistant depression and generating a corresponding ranking of brain region importance.

Claims

1. A brain imaging system for treatment-resistant depression, comprising a plurality of hardware modules composed of a plurality of hardware circuits, the hardware modules including: A communication module is connected to a hospital host computer; an image processing module is connected to the communication module. After receiving a FDG-PET medical image of a patient transmitted by the hospital host computer through the communication module, the module converts the format of the FDG-PET medical image, aligns the converted FDG-PET medical image to a standard brain template for spatial normalization, and corrects the normalized FDG-PET medical image using a standardized uptake value to generate a preprocessed medical image. The image processing module then divides the preprocessed medical image into multiple brain regions using an Automated Anatomical Labeling (AAL) template, and standardizes the preprocessed medical image based on the average metabolic value of each brain region to generate a standardized preprocessed medical image. A machine learning module, communicatively connected to the image processing module and the communication module, receives standardized preprocessed medical images and uses multiple machine learning models to analyze the standardized preprocessed medical images for treatment-resistant depression based on brain glucose metabolism characteristics to obtain a treatment-resistant depression diagnosis result. After receiving a Hannspree Syndrome Scale (HSS) of the patient transmitted from the hospital host via the communication module, the machine learning module uses these machine learning models to analyze the HSS to assess the severity of the patient's depression and generate a depression severity diagnosis result. A feature analysis module, communicatively connected to the machine learning module and the communication module, receives the treatment-resistant depression diagnosis result and the depression severity diagnosis result, and performs an influence analysis on each brain region in the treatment-resistant depression diagnosis result to analyze the individual influence of each brain region in each machine learning model on treatment-resistant depression and generate a corresponding brain region importance ranking.

2. The brain imaging system for treatment-resistant depression as described in claim 1, wherein the machine learning models are used to perform image processing calculations on models including logistic regression, support vector machine, multilayer perceptron, random forest, CatBoost, XGBoost, or LightGBM based on the brain glucose metabolism rate characteristics.

3. The brain imaging system for treatment-resistant depression as described in claim 1, wherein the format conversion is performed to convert the positron emission tomography medical image from a DICOM file to an NIfTI format.

4. The brain imaging processing system for treatment-resistant depression as described in claim 1, wherein the method for influence analysis is SHAP analysis.

5. The brain imaging system for treatment-resistant depression as described in claim 1 further includes a user interface communicatively connected to the communication module, for allowing the user to input patient information and display the diagnosis results of treatment-resistant depression, the diagnosis results of depression severity, and the ranking of the importance of the brain regions.

6. A brain imaging processing method for treatment-resistant depression, comprising the following steps: An image processing module receives an positron emission tomography (PET) image of a patient transmitted from a hospital host via a communication module, and converts the PET image into a new format; The image processing module aligns the converted PET image to a standard brain template for spatial normalization, and corrects the normalized PET image using a standardized uptake value to generate a preprocessed medical image; The image processing module divides the preprocessed medical image into multiple brain regions using an automatic anatomical annotation template, and standardizes the preprocessed medical image using the average metabolic value of each brain region as a benchmark to generate a standardized preprocessed medical image; A machine learning module uses multiple machine learning models to analyze the standardized preprocessed medical image for treatment-resistant depression based on brain glucose metabolism characteristics to obtain a diagnosis of treatment-resistant depression. The machine learning module also receives a Hannspree Depression Scale from the patient, transmitted by the hospital host through the communication module, and uses these machine learning models to analyze the Hannspree Depression Scale to assess the severity of the patient's depression and generate a depression severity judgment result; and a feature analysis module performs an influence analysis on each brain region of the treatment-resistant depression judgment result and the depression severity judgment result to analyze the individual influence of each brain region in each machine learning model on treatment-resistant depression and generate a corresponding brain region importance ranking.

7. The brain imaging processing method for treatment-resistant depression as described in claim 6, wherein the format conversion is to convert the medical image from a DICOM file to an NIfTI format.

8. The brain image processing method for treatment-resistant depression as described in claim 6, wherein the machine learning models are used to perform image processing calculations on models including logistic regression, support vector machine, multilayer perceptron, random forest, CatBoost, XGBoost or LightGBM based on the brain glucose metabolism rate characteristics.

Images