A diagnostic and prognostic prediction method for mental health issues

Abstract

The present invention relates to a method for diagnostic and prognostic prediction of mental issues based on neuroimages using deep learning algorithms. More particularly, the present invention relates a method for classification and early diagnosis of mental issues based on neuroimages like electroencephalography (EEG) signals or magnetic resonance images (MRI) by using deep learning algorithms with high accuracy.

Description

[0001]DESCRIPTION A DIAGNOSTIC AND PROGNOSTIC PREDICTION METHOD FOR MENTAL HEALTH ISSUES Technical Field of the Invention: The present invention relates to a method for diagnostic and prognostic prediction of mental issues based on neuroimages using deep learning algorithms. More particularly, the present invention relates a method for classification and early diagnosis of mental issues based on neuroimages like electroencephalography (EEG) signals or magnetic resonance images (MRI) by using deep learning algorithms with high accuracy. Background of the Invention: Disorders that affect people's thoughts, feelings, and behaviors and are associated with distress or impairment in important areas of functionality are called mental health disorders. Mental health disorder is a broader term that covers mental disorders, psychosocial disabilities, and mental states at risk from dysfunction to self-harm. Mental health disorders can be divided into 7 main groups: anxiety disorders, mood disorders, psychotic disorders, eating disorders, personality disorders, dementia, and autism. Anxiety disorders include generalized anxiety disorder, panic disorder, obsessive-compulsive disorder (OCD), post- traumatic stress disorder (PTSD), and social anxiety disorder. Mood disorders include depression, bipolar disorder, and dysthymic disorder. Psychotic disorders include schizophrenia, schizoaffective disorder, delusional disorder, and brief psychotic disorder. Eating disorders include anorexia nervosa, bulimia nervosa, binge eating disorder, pica eating disorder, and rumination disorder. Personality disorders include borderline personality disorder, narcissistic personality disorder, antisocial personality disorder, and dissociative identity disorder. Dementia includes Alzheimer’s disease, Parkinson’s disease, Huntington’s disease, and Wernicke-Korsakoff syndrome. Accurate and early diagnosis is very important for the right treatment of these kinds of disorders. Although no medical diagnostic test exists, mental health disorders are diagnosed with a physical exam and long-term monitoring by healthcare professionals like doctors, psychiatrists, or psychologists. In order to properly evaluate a patient’s condition, professionals use neuroimaging techniques. Three different neuroimaging techniques, electroencephalogram (EEG), magnetic resonance imaging (MRI), and positron emission tomography (PET) help to explore and measure brain activity. Although image results of the neuroimaging techniques are used to detect the existing mental health disorder in the patient, image results of the neuroimaging techniques cannot be used to early diagnose the mental health disorder or to determine the patient's susceptibility to the mental health disorder. Today, imaging results of neuroimaging techniques are processed with machine learning algorithms to diagnose and predict delicacy early. Deep learning models that are a subclass of the machine learning algorithms are used the neuroimaging techniques results as input data to extract features for the diagnosing process. Deep learning models have several limitations to extract features from data and making a decision. First of all, deep learning models suffer from a lack of interpretability to make decisions from the data. In order to make a decision, deep learning models need training with large-size data. Besides, after training deep learning models become effective on specialized data types to make a decision. Therefore, deep learning models need training for different data types since their overfitting and transferability abilities are poor. In addition, training a deep learning model requires a large amount of data, power, and time. Deep learning models are susceptible to adversarial attacks, where a small perturbation to the input data can cause the model to make incorrect predictions. Therefore, deep learning models also train against adversarial attacks. Deep learning models can identify correlations between variables however they can not establish causality. Moreover, deep learning models lack the ability to reason and understand common sense concepts, which limits their ability to perform tasks that require high-level reasoning and understanding. In the state of the art, several patent applications are disclosing different methods to early diagnose and predict mental issues based on neuroimages using deep learning algorithms and some of them are discussed below. The document with publication number "US2014303424A1" and named “METHODS AND SYSTEMS FOR DIAGNOSIS AND TREATMENT OF NEURAL DISEASES AND DISORDERS” is related to methods and systems for modulating activity of a nervous system component. Neural pattern recognition is used to identify a neurological and / or psychiatric disease or disorder based on input generated by electric signals indicative of the patient's brain activity. The method of the said document used fMRI, which is an expensive and time-consuming imaging technique that requires specialized equipment. Besides, the method of the said document used a conventional machine learning method, which may not be as effective as deep learning in handling complex and high-dimensional data. Furthermore, the method of the said document does not use preprocessing to remove the contaminations from data which reduces the accuracy of the diagnosis and prediction. Moreover, the method of the said document does not explain the evaluation process that ensures the reliability and validity of the diagnosis and prediction. The document with publication number "US10758174B2" and named “Method for cross-diagnostic identification and treatment of neurologic features underpinning mental and emotional disorders” is related to a system and method for diagnosing mental or emotional disorders. An effective brain- computer interface component is incorporated into a closed-loop, symptom—responsive psychiatric deep-brain stimulation system and the system identifies what is abnormal for that individual patient in terms of functional domains by examining the patient's performance on the task battery. The method of the said document used and compared instantaneous measured values over various activities and psycho-physical task results to diagnose. These prior art documents do not solve the problems required to early diagnose and predict delicacy by using deep learning methods. Existing diagnosis and prediction methods for mental health disorders have some drawbacks like subjectivity, reliability, expense, accessibility, stigma, large diagnostic criteria, and longitudinal data. As a result, due to the shortcomings described above and the inadequacy of the existing solutions on the subject, it was necessary to make an improvement in the relevant technical field. Objectives of the Invention The main object of the invention is to early diagnose and predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques. For this reason, electroencephalogram (EEG) or magnetic resonance imaging (MRI) results of the patients are used for detection, prediction, and classification for early diagnosis and prognostic prediction by using a deep learning model. Another object of the invention is to provide a clinically useful method by using the raw form of basic neuroimaging techniques results such as EEG and MRI to extract features related to diseases, classify diseases, and express classification performance with parameters such as ROC curve, AUC value, and high accuracy. Consequently, the method ensures the reliability and validity of the diagnosis and prediction. Another object of the invention is to use multi-modal data like EEG results, psychiatric interviews, neuropsychological testing, and treatment results, from both unhealthy and healthy control groups. In this way, the method allows for a more comprehensive understanding of mental health disorders and can help improve the accuracy and reliability of the predictions. Another object of the invention is to use deep learning algorithms with trained multi-model data to analyze neuroimaging data and make predictions. For this reason, the method improves the accuracy and reliability of mental health diagnosis and prediction. Another object of the invention is to evaluate the prediction accuracy using external validation data, increasing the reliability and generalizability of the method. In this way, the method ensures effectiveness for a diverse range of patients and is not limited to the data used to train the model. Another object of the invention is to include converting the EEG signals to one-dimensional waveform data by deleting contaminations on EEG segments. In this way, the method ensures higher accuracy in the conversion process. Another object of the invention is to visualize a topographical map with a feature heat map to extract temporal and spatial features related to diseases. For this reason, the method provides accurate results to diagnose and predict mental health disorders. Another object of the invention is to use EEG signals, which are non-invasive and can be easily obtained from patients. In this way, the method of invention eliminates the need to use expensive and time- consuming equipment. Another object of the invention is to employ deep learning techniques to extract temporal and spatial features from EEG signals, which can accurately diagnose and predict mental health disorders. For this reason, extracting temporal and spatial features from handling complex and high-dimensional data is effective. Another object of the invention is to include a manual step for data preprocessing by removing contaminations from EEG segments, which can improve the accuracy of the diagnosis and prediction. Another object of the invention is to propose a web interface for clinicians to upload raw data to display the classification of the patient and the treatment response. In his way, the treatment response can help improve patient outcomes and represents a shift toward more individualized and patient- centered care. Brief Description of the Drawings: Figure 1; Schematic illustration of a flow diagram of the process according to the present invention. Reference Numerals: 100. Creating a model 110. Collecting neuroimaging data that are EEG or MRI data, psychiatric interviews, neuropsychological testing, treatment results, and various scales from patients who suffer from various mental health disorders already diagnosed by psychiatrists as an unhealthy group 120. Collecting neuroimaging data that are EEG or MRI data, psychiatric interviews, neuropsychological testing, and various scales are collected from patients who were not already diagnosed by psychiatrists as a healthy control group 130. Recording the unhealthy group and the healthy control group in a database 140. Training the model with recorded the unhealthy group and the healthy control group data 200. Collecting neuroimaging data that are EEG or MRI data from a patient who applied for a neuropsychiatric examination from a user interface to the server 300. Preprocessing of the neuroimaging data 310. Filtering the collected EEG data from 0.1 to 50 Hz to remove the artifacts 311. Analyzing the filtered EEG data to determine contaminations that are related to eye movement 312. Converting the EEG data to one-dimensional waveform data by deleting the contaminations on EEG segments 320. Spatially normalizing and segmenting the collected MRI data into gray and white matter, and cerebrospinal fluid (CSF) by applying image denoising, bias correction, and intensity normalization to prepare the collected MRI data for a segmentation 321. Segmenting the MRI data into white matter, grey matter, and cerebrospinal fluid 322. Identifying and segmenting blood vessels within the MRI data by applying ROI segmentation 323. Refining the MRI data by applying blood vessel correction, optimized shooting registration, and final cleanup 324. Controlling the MRI data based on noise, resolution, and bias field correction 400. Extracting discriminative and informative features from the neuroimaging data 500. Comparing the extracted features with a database 600. Evaluate the comparison results with parameters that are accuracy, sensitivity, specificity, kappa value, receiver operating characteristic (ROC) curve with AUC value, and external validation data accuracy 700. Visualizing the evaluation result as a clinical result 800. Determining treatment options according to the clinical result based on treatment result information that is recorded in the database 900. Transmitting the clinical result and treatment options to the user interface Detailed Description of the Invention: The present invention relates to a diagnostic and prognostic prediction method that is run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method. The server is a cloud server. Therefore, any kind of device that has the ability to connect to the Internet and a user interface can connect with the cloud server since the server is not stable upon location. The method of the present invention comprises 4 main process steps. The first step is neuroimaging data collection from the patients suffering from a specific psychiatric disease. The collected neuroimaging data could be EEG or MRI data. If the neuroimaging data is an EEG signal, the EEG signal acquisition from a patient who applied for a neuropsychiatric examination. The EEG signal is captured in a room with low lighting, a temperature range of 21 to 26°C, and a humidity range of 30 to 60%. The patient is seated in a chair with 120 degrees tilt, the patient’s forearms are rested on the armrests to ensure a comfortable posture, and the patient is wore earplugs to block out background noise. In addition, the patient is instructed to avoid excessive eye-blinking and motion while the EEG signals are recorded. EEG signals are captured in the eyes-closed condition for a total of three minutes. International 10 / 20 nomenclature is used to record a 19-channel EEG with active electrodes embedded in a cap. The electrode impedance is maintained below 5 KΩ. Using Scan LT (version 1.2), EEG signals are captured at 125 Hz with the ground electrode positioned halfway between the Fpz and FZ electrodes. The captured EEG signals are collected from a user interface to the server. Electroencephalogram (EEG) results, psychiatric interviews, neuropsychological testing, treatment results, and various scales are collected from patients who suffer from various mental health disorders already diagnosed by psychiatrists as an unhealthy group. For a healthy control group, electroencephalogram (EEG) results, psychiatric interviews, neuropsychological testing, and various scales are collected from patients who were not already diagnosed by psychiatrists. Then, the data of the unhealthy group and the healthy control group are recorded in a database for comparison. During data collection, neurological disorders, head injuries, and drug or alcohol consumption are not included as diagnostic criteria. If the neuroimaging data is an MRI image, the MRI image is captured in an MRI machine with high-resolution T1-weighted images acquired using the 3D magnetization- prepared rapid acquisition gradient echo (MPRAGE) technique. High-resolution pictures of the brain are produced using a T1-weighted gradient echo sequence using this three-dimensional (3D) structural imaging method. By rapidly sending a sequence of gradient magnetic field pulses to the tissue of interest while it is magnetized, the MPRAGE technique causes the tissue to release a signal that the scanner can detect. During capturing the MRI images, the patient is laid down in the MRI machine. Besides, the patient is instructed to avoid excessive motion while the MRI images are recorded. The captured MRI images are collected from a user interface to the server. MRI (Magnetic Resonance Imaging) results, psychiatric interviews, neuropsychological testing, treatment results, and various scales are collected from patients who suffer from various mental health disorders already diagnosed by psychiatrists as an unhealthy group. For a healthy control group, MRI results, psychiatric interviews, neuropsychological testing, and various scales are collected from patients who were not already diagnosed by psychiatrists. Then, the data of the unhealthy group and the healthy control group are recorded in a database for comparison. During data collection, neurological disorders, head injuries, and drug or alcohol consumption are not included as diagnostic criteria. The second step is pre-processing run on the server. If the neuroimaging data is an EEG signal, pre- processing step comprises artifact removal, filtering, and independent component analysis steps. In order to remove the artifacts from the captured EEG signals, the EEG signals are filtered from 0.1 to 50 Hz. This range is considered less rigid in studying neuropsychiatric diseases and more informative for deep learning processing. After filtering, the EEG signals are analyzed to determine contaminations that are related to eye movement, and then to eliminate these contaminations, they are manually or automatically deleted on the EEG segments. After filtering, to determine and eliminate the contaminations bandpass filtering, independent component analysis (ICA), regression-based methods, wavelet transform, and notch filter methods are used individually or in combination. The bandpass filtering method is used to filter the EEG signal within specific frequency bands and can help remove eye movement artifacts. Eye movement artifacts are typically low-frequency signals. Therefore, a high- pass filter can be applied to remove any DC drift, followed by a bandpass filter between 1-20 Hz. The independent component analysis is a signal processing technique that can separate EEG signals into independent components. Since eye movements produce unique artifacts, the components related to eye movements are identified and removed from the signal by using the independent component analysis technique. The regression-based methods involve creating a model that explains the relationship between the EEG signal and the eye movement artifacts. After a model is created, the regression-based method is used to remove the eye movement artifacts from the EEG signal. Wavelet transform is a mathematical technique that can decompose the EEG signal into different frequency components. The frequency components that are related to eye movement artifacts are identified and removed from the signal by using the wavelet transform techniques. The remained bad fragments were then manually extracted. The first 180 seconds of the EEG data is cropped and exported for further analysis of the process. If the neuroimaging data is MRI, after the MRI image acquisition, the pre-processing step comprises format conversion, noise reduction, image registration, image segmentation, bias field correction, skull stripping, normalization, and quality control. The captured MRI images are usually saved in DICOM format and the images should be converted to NIfTI format for analysis. To increase the accuracy of the analysis, Gaussian filtering or wavelet denoising is used in the noise reduction step. In the image registration step, multiple images of the same patient are aligned for comparing images acquired at different times or by different scanners. Image segmentation is the process of dividing the image into different regions based on their properties for identifying specific structures or regions of interest in the image. The bias field correction is used to correct the inhomogeneity in the intensity of the MRI images caused by variations in the radio frequency field by using algorithms such as N3 or FSL-BET to correct the bias field. To analyze the brain structure, the skull is removed from captured MRI images with using automated algorithms or manual methods. The normalization step involves scaling the intensity values of the captured MRI images to a common range for comparing images acquired from different patients or scanners. The quality control step which is the last step of the preprocessing involves checking the MRI images for artifacts, motion, and other anomalies that can affect the accuracy of the analysis. These preprocessing steps are used for ensuring the accuracy and reliability of the MRI analysis. For preprocessing of captured MRI images are spatially normalized, and segmented into gray and white matter, and cerebrospinal fluid (CSF) by applying image denoising, bias correction, and intensity normalization to prepare the data for the segmentation. In the segmentation step, the MRI images are segmented into white matter, grey matter, and cerebrospinal fluid. After the segmentation step, ROI segmentation is applied to identify and segment blood vessels within the MRI images. Finally, the segmentation is refined by applying blood vessel correction, optimized shooting registration, and final cleanup. To make sure that the preprocessed data is appropriate for further analysis, the preprocessed MRI images are controlled based on the three parameters: noise, resolution, and bias. The overall threshold above 80% ensures a high image and preprocessing quality. The third step comprises the server's feature extraction, classification, and evaluation process steps. For EEG data, one commonly used deep learning model for classification is the Convolutional Neural Network (CNN). CNNs have been shown to be effective in capturing the spatial and temporal features present in EEG signals. The input to the CNN would be a sequence of EEG recordings, and the output would be a classification label indicating the presence or absence of a particular psychiatric disease. A possible CNN architecture for EEG data classification would consist of a series of convolutional layers with pooling layers in between, followed by fully connected layers. The convolutional layers would learn different filters to detect relevant features in the EEG signals, while the pooling layers would reduce the dimensionality of the output of the convolutional layers. The fully connected layers at the end of the network would then perform the classification based on the extracted features. A 1D Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) layers is another deep learning model for EEG classification. This model is particularly useful for capturing both the temporal and spatial features of the EEG data. The input to the model would be a sequence of EEG recordings, where each recording represents a fixed time interval (e.g., 1 second). The output would be a classification label indicating the presence or absence of a particular psychiatric disease. The first layer of the model would be a 1D convolutional layer, which would learn filters to detect different temporal patterns in the EEG signals. The output of this layer would then be fed into an LSTM layer, which is a type of recurrent neural network that can remember long-term dependencies in sequential data. The LSTM layer would capture the temporal dynamics of the EEG data, and its output would be fed into a series of fully connected layers for classification. The 1D CNN+LSTM architecture allows the model to capture both the short-term and long-term temporal patterns in the EEG data. The CNN layer can identify short-term patterns in the EEG signal, such as sharp waveforms or high-frequency oscillations, while the LSTM layer can capture longer-term patterns, such as slow-wave activity or changes in EEG amplitude over time. The model can be optimized during training using backpropagation and stochastic gradient descent to minimize a cross-entropy loss function. Regularization techniques such as dropout or weight decay can also be applied to prevent overfitting. Overall, the 1D CNN+LSTM model is a powerful deep learning approach for EEG classification, as it can effectively capture both the temporal and spatial features of EEG data. For structural MRI data, one possible deep learning model for classification is the 3D Convolutional Neural Network (3D CNN).3D CNNs are similar to traditional 2D CNNs but operate on volumetric data. They have been shown to be effective in capturing the 3D spatial patterns in structural MRI data. The input to the 3D CNN would be a 3D structural MRI scan, and the output would be a classification label indicating the presence or absence of a particular psychiatric disease. The 3D CNN architecture would consist of a series of 3D convolutional layers with pooling layers in between, followed by fully connected layers for classification. Similar to the EEG model, the convolutional layers would learn different filters to detect relevant features in the structural MRI scans, while the pooling layers would reduce the dimensionality of the output of the convolutional layers. The fully connected layers at the end of the network would then perform the classification based on the extracted features. Both the EEG and structural MRI models can be trained using backpropagation and stochastic gradient descent to minimize a cross-entropy loss function. The models can also be regularized using techniques such as dropout or weight decay to prevent overfitting. Additionally, transfer learning can be used to leverage pre-trained models on similar tasks or datasets, to improve performance with a smaller set of training data. The present invention's model is run the one-dimensional waveform EEG data to extract temporal and spatial features and compare them with the data recorded in the first database for classification. During the comparison, the model used parameters that are accuracy, sensitivity, specificity, kappa value, receiver operating characteristic (ROC) curve with AUC value, and external validation data accuracy for evaluation. Thus, the evaluation step is important to measure the robustness, interpretability, and reliability of the method. Extracting temporal and spatial features is essential to diagnosis and prognostic prediction. Feature extraction is necessary to reduce the dimensionality of EEG signals, enhance the signal-to-noise ratio, identify patterns and relationships, and improve classification and prediction accuracy. By extracting relevant features from EEG signals, it is possible to gain insights into the neural mechanisms underlying cognitive, sensory, and motor processes and diagnose and monitor neurological disorders. Convolutional neural networks, recurrent neural networks, autoencoders, and generative adversarial networks are used to automatically extract features from EEG signals, allowing for more efficient and accurate analysis of large amounts of data. These methods are also used for a variety of tasks such as classification, regression, and anomaly detection in EEG data. As to the feature extraction process for MRI data, the magnetic characteristics of hydrogen molecules within tissues are reflected in the MRI pictures, which are rich in texture and enable feature extraction. A feature or pattern that appears repeatedly on an image's surface and symbolizes its structural elements is said to have a certain texture. The goal of texture analysis is to identify a distinctive method for simplification and expression of the hidden properties of textures. Since the changes in gray matter tissue that are indicative of the specified disorders are of relevance to us, we used preprocessed gray matter volumes. The last step is the user interface design user interface to interact with the user. Since there are multiple models for each psychiatric disorder the user is free to choose any model for external validation and display the prediction result for each model. In addition to the diagnosis process, the prognostic model interface will be displaying the potential treatment options and treatment response for the diseases The diagnostic and prognostic prediction method for mental health disorders that run on a server comprises the following process steps for neuroimaging data; • Creating a model (100), o Collecting neuroimaging data that are EEG or MRI data, psychiatric interviews, neuropsychological testing, treatment results, and various scales from patients who suffer from various mental health disorders already diagnosed by psychiatrists as an unhealthy group (110), o Collecting neuroimaging data that are EEG or MRI data, psychiatric interviews, neuropsychological testing, and various scales are collected from patients who were not already diagnosed by psychiatrists as a healthy control group (120), o Recording the unhealthy group and the healthy control group in a database (130), o Training the model with recorded the unhealthy group and the healthy control group data (140), • Collecting neuroimaging data that are EEG or MRI data from a patient who applied for a neuropsychiatric examination from a user interface to the server (200), • Preprocessing of the neuroimaging data (300), • Extracting discriminative and informative features from the neuroimaging data (400), • Comparing the extracted features with a database (500), • Evaluate the comparison results with parameters that are accuracy, sensitivity, specificity, kappa value, receiver operating characteristic (ROC) curve with AUC value, and external validation data accuracy (600), • Visualizing the evaluation result as a clinical result (700), • Determining treatment options according to the clinical result based on treatment result information that is recorded in the database (800), • Transmitting the clinical result and treatment options to the user interface (900). If the collected neuroimaging data is EEG data in the process step of preprocessing comprises the following process steps; ▪ Filtering the collected EEG data from 0.1 to 50 Hz to remove the artifacts (310), ▪ Analyzing the filtered EEG data to determine contaminations that are related to eye movement (311), ▪ Converting the EEG data to one-dimensional waveform data by deleting the contaminations on EEG segments (312). If the collected neuroimaging data is MRI data in the process step of preprocessing comprises the following process steps; ❖ Spatially normalizing and segmenting the collected MRI data into gray and white matter, and cerebrospinal fluid (CSF) by applying image denoising, bias correction, and intensity normalization to prepare the collected MRI data for a segmentation (320), ❖ Segmenting the MRI data into white matter, grey matter, and cerebrospinal fluid (321), ❖ Identifying and segmenting blood vessels within the MRI data by applying ROI segmentation (322), ❖ Refining the MRI data by applying blood vessel correction, optimized shooting registration, and final cleanup (323), ❖ Controlling the MRI data based on noise, resolution, and bias field correction (324).

Claims

CLAIMS 1. A diagnostic and prognostic prediction method for mental health disorders that are run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method, characterized by comprising steps; ▪ Creating a model (100), ▪ Collecting neuroimaging data that are EEG or MRI data from a patient who applied for a neuropsychiatric examination from a user interface to the server (200), • Preprocessing of the neuroimaging data (300), • Extracting discriminative and informative features from the neuroimaging data (400), • Comparing the extracted features with a database (500), • Evaluate the comparison results with parameters that are accuracy, sensitivity, specificity, kappa value, receiver operating characteristic (ROC) curve with AUC value, and external validation data accuracy (600), • Visualizing the evaluation result as a clinical result (700), • Determining treatment options according to the clinical result based on treatment result information that is recorded in the database (800), • Transmitting the clinical result and treatment options to the user interface (900).

2. A diagnostic and prognostic prediction method for mental health disorders that are run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1, in the process step of creating a model (100) comprises; a. Collecting EEG, MRI data, psychiatric interviews, neuropsychological testing, treatment results, and various scales from patients who suffer from various mental health disorders already diagnosed by psychiatrists as an unhealthy group (110), b. Collecting EEG and MRI data, psychiatric interviews, neuropsychological testing, and various scales are collected from patients who were not already diagnosed by psychiatrists as a healthy control group (120), c. Recording the unhealthy group and the healthy control group in a database (130), d. Training the model with recorded the unhealthy group and the healthy control group data (140).

3. A diagnostic and prognostic prediction method for mental health disorders that are run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1, if the collected the neuroimaging data is EGG data in the process step of preprocessing of the neuroimaging data (300) comprises; ▪ Filtering the collected EEG data from 0.1 to 50 Hz to remove the artifacts (310), ▪ Analyzing the filtered EEG data to determine contaminations that are related to eye movement (311), ▪ Converting the EEG data to one-dimensional waveform data by deleting the contaminations on EEG segments (312).

4. A diagnostic and prognostic prediction method for mental health disorders that are run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1, if the collected the neuroimaging data is MRI data in the process step of preprocessing of the neuroimaging data (300) comprises; ❖ Spatially normalizing and segmenting the collected MRI data into gray and white matter, and cerebrospinal fluid (CSF) by applying image denoising, bias correction, and intensity normalization to prepare the collected MRI data for a segmentation (320), ❖ Segmenting the MRI data into white matter, grey matter, and cerebrospinal fluid (321), ❖ Identifying and segmenting blood vessels within the MRI data by applying ROI segmentation (322), ❖ Refining the MRI data by applying blood vessel correction, optimized shooting registration, and final cleanup (323), ❖ Controlling the MRI data based on noise, resolution, and bias field correction (324).

5. A diagnostic and prognostic prediction method for mental health disorders that are run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1 and Claim 3, in the process step of converting the EEG data to one-dimensional waveform data by deleting the contaminations on EEG segments (312), wherein deleting the contaminations on EEG segments is done manually.

6. A diagnostic and prognostic prediction method for mental health disorders that is run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1 and Claim 3, in the process step of analyzing the filtered EEG data to determine contaminations that are related to eye movement (311), wherein determining the contaminations bandpass filtering, independent component analysis (ICA), regression-based methods, wavelet transform, and notch filter methods are used individually or in combination.

7. A diagnostic and prognostic prediction method for mental health disorders that is run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1, if preprocessed neuroimaging data is EGG data in the process step of extracting discriminative and informative features from the neuroimaging data (400), wherein extracting discriminative and informative features are temporal and spatial features of the one-dimensional waveform EEG data.

8. A diagnostic and prognostic prediction method for mental health disorders that is run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1 and Claim 7, if preprocessed neuroimaging data is EGG data in the process step of extracting discriminative and informative features from the neuroimaging data (400), wherein extracting the temporal and spatial features of the one- dimensional waveform EEG data convolutional neural networks, recurrent neural networks, autoencoders, and generative adversarial networks are used.

9. A diagnostic and prognostic prediction method for mental health disorders that is run on a server to early diagnose or predict whether a patient will have mental health disorders in the future or not from a patient’s image results of the neuroimaging techniques by using the deep learning method according to Claim 1, if preprocessed neuroimaging data is MRI data in the process step of extracting discriminative and informative features from the neuroimaging data (400), wherein extracting discriminative and informative features are magnetic characteristics of hydrogen molecules within tissues are reflected in the MRI data.

Images