ADHD based on deep network and subtype diagnosis method thereof

Abstract

The invention discloses an ADHD and subtype diagnosis method based on a deep network. The ADHD and subtype diagnosis method based on the deep network comprises the following steps: establishing a sight tracking network CognittionNet; the method comprises the following steps: acquiring an attention signal At (t) and an iris signal Ir (t) by using a front short video input CognittionNet of a person to be detected; the variance (Atstd) and the average change speed (Atvmean) of At (t) are counted; counting the average change speed (Irvmean) of the Ir (t); judging whether the person to be detected is ADHD or non-ADHD; and judging and confirming whether the ADHD patient belongs to an attention deficit type (ADHD-I), a hyperactivity impulse type (ADHD-II) or a hybrid type (ADHD-III). Therefore, the objective diagnosis of the ADHD and the subtype thereof can be realized by only adopting one short video.

Description

Technical Field

[0001] This invention belongs to the field of deep learning, specifically a method for diagnosing ADHD and its subtypes based on deep networks. Background Technology

[0002] Attention deficit hyperactivity disorder (ADHD) is a common neurodevelopmental disorder that begins in childhood and can persist into adulthood. According to the DSM-V report, the prevalence of ADHD is 5% in most cultures worldwide, and approximately 5.7% in China.

[0003] Currently, the diagnosis of ADHD mainly relies on subjective methods such as clinical interviews and behavioral scales, and there are no clear biomarkers. In recent years, functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and eye tracking have been widely used in objective identification research of ADHD. fMRI combined with graph convolutional neural networks (GCN) can achieve a classification accuracy of up to 95.2%, but this method has problems such as complex data processing, sensitivity to head movement artifacts, and insufficient model interpretability. EEG often uses the theta / β wave power ratio as the main feature, and some studies have reported an accuracy of up to 99% (Mandal, 2024), but its specificity and cross-individual consistency still need further verification. In contrast, eye tracking data is receiving increasing attention due to its non-invasiveness and ease of acquisition. Wainstein et al. (2017) found that dynamic changes in pupils can effectively distinguish individuals with ADHD; Ko et al. (2022) achieved an accuracy of over 82% based on the Mel frequency cepstral coefficient (MFCC) feature of eye movements; Yoo et al. (2024) identified 33 key eye movement features closely related to cognitive function, further confirming their application prospects in the identification of mental illnesses.

[0004] Traditional eye tracking (such as infrared reflection-based devices) typically requires head fixation and has limited accuracy. In recent years, deep neural networks (DNNs) have shown significant advantages in gaze estimation. The MPIIGaze model proposed by Zhang et al. reduced the gaze estimation error to 5.4°; Xucong Zhang (2017) further introduced a full-face attention mechanism to improve estimation accuracy. However, these methods are still susceptible to changes in lighting and head posture, and their performance is still lower than that of eye trackers.

[0005] To overcome the aforementioned problems, this patent establishes a novel deep neural network architecture—Cognition_Net. This network structure achieves high-precision eye movement feature extraction by focusing on key regions in the face related to eye movements, and its performance is comparable to that of professional-grade eye trackers.

[0006] Cognition_Net analyzes short, natural video sequences to extract three physiologically specific indicators: attention, average eye movement velocity, and average pupil size change velocity. Based on these three indicators, this patented technology can distinguish between children with ADHD and healthy children with over 99% accuracy, and can also differentiate between the three subtypes of ADHD, without the need for specialized equipment or specific experimental conditions. Attached Figure Description

[0007] Figure 1 This is a flowchart of the patent; Figure 2 This is a diagram of the detection system. Figure 3 This describes the overall structure of the eye-tracking network. Figure 4 To introduce a wavelet attention feature fusion module (WA); Figure 5 Iris and attentional signals in ADHD patients; Figure 6 This represents the regional distribution of various samples within the coordinates established by the index. Detailed Implementation

[0008] The purpose of this invention is to implement a deep network-based diagnostic method for ADHD and its subtypes using a short video clip. The invention will now be described in detail with reference to the accompanying drawings.

[0009] A deep network-based diagnostic method for ADHD and its subtypes, the process is as follows: Figure 1 As shown, it includes the following steps: Step 1: Establish the gaze-tracking network Cognition_Net; Step 2: Input the short frontal video of the subject into Cognition_Net to obtain attention signals. At(t) and iris signal Ir(t) ; Step 3, Statistics At(t) variance ( At_std ), average rate of change ( At_v_mean );statistics Ir(t) average rate of change ( Ir_v_mean ); Step four: Determine whether the person being tested has ADHD or not; Step 5: Determine whether the diagnosed ADHD patient belongs to the Attention Deficit Hyperactivity Disorder (ADHD-I), the Hyperactivity-Impulsive Disorder (ADHD-II), or the Mixed Disorder (ADHD-III).

[0010] The diagnostic system's hardware and software, such as Figure 2As shown, the setup includes a PC and a webcam. The PC hardware consists of an Intel(R) Core(TM) i9-9900K, an NVIDIA GTX1080Ti, 16GB of DDR4 RAM, and a 500GB hard drive; the software environment is an Ubuntu 16.04 operating system, Python 3.7.4, and the PyTorch 1.4.0 deep learning framework. The webcam operates at a frame rate of 60fps with a resolution of 1280*720.

[0011] The camera was placed on the top edge of the computer monitor, and the participants looked at the screen from a distance of about 60cm, in a natural sitting posture.

[0012] The following explains the specific implementation of the four steps: Step one involves building a lightweight eye-tracking network, Cognition_Net, which specifically involves the following two steps: 1) By embedding a multi-scale feature fusion module (EfficientMLCA) and a bidirectional long short-term memory network (Bi-LSTM) into the mid-to-high-level feature outputs of the original backbone network ResNet18, and introducing a wavelet attention feature fusion module (WA), a Cognition_Net network is constructed. The overall network structure is as follows: Figure 3 As shown, a wavelet attention feature fusion module (WA) is introduced, as follows: Figure 4 As shown; 2) The Cognition_Net network was trained using the ETH-XGaze dataset.

[0013] In step two, attention signals are obtained by inputting a short frontal video of the subject into Cognition_Net. At(t) and iris signal Ir(t) The specific steps are as follows: 1) Collect a short frontal video of the person to be examined. I(t) ; 2) Short videos are input into Cognition_Net to obtain attention signals. At(t) and iris signal Ir(t) .

[0014] Statistics in step three At(t) variance ( At_std ), average rate of change ( At_v_mean );statistics Ir(t) average rate of change ( Ir_v_mean The specific steps are as follows: 1) Attention signals At(t) Perform Kalman filtering and calculate At(t) of At_std , At_v_mean Its time-domain waveform is as follows Figure 5As shown.

[0015] 2) Iris signal Ir(t) Perform wavelet filtering and calculate Ir(t) of Ir_v_mean Its time-domain waveform is as follows Figure 5 As shown.

[0016] Step four determines whether the person being tested has ADHD or not, using the following methods: If 0 < At_std < 0.14, 0 < At_v_mean < 4.05, 0 < Ir_v_mean < 1230, The child is a normal child, such as Figure 6 If the child is located in a rectangular area near the original coordinate axis, then the child has ADHD.

[0017] Step 5: Determine whether the diagnosed ADHD patient belongs to the Attention Deficit Hyperactivity Disorder (ADHD-I), the Hyperactivity-Impulsive Disorder (ADHD-II), or the Mixed Disorder (ADHD-III). The specific steps are as follows: 1) If At_std > 0.14, 0 < At_v_mean < 4.05, Ir_v_mean > 1230, The child has ADHD-I, such as Figure 6 The rectangular area located to the right of the non-ADHD rectangle.

[0018] 2) If 0 < At_std < 0.14, At_v_mean > 4.05, Ir_v_mean > 1230, The child has ADHD-II; such as Figure 6 The rectangular area located behind the non-ADHD rectangle.

[0019] 3) If [[ID= The child has ADHD-III. ​ A region outside the aforementioned areas.

[0020] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing ADHD and its subtypes based on deep networks, characterized in that, Includes the following steps: Step 1: Establish the gaze-tracking network Cognition_Net; Step 2: Input the short frontal video of the subject into Cognition_Net to obtain attention signals. At(t) and iris signal Ir (t) ; Step 3, Statistics At(t) variance At_std Average rate of change At_v_mean ,statistics Ir(t) average rate of change Ir_v_mean ; Step four: Determine whether the person being tested has ADHD or not; Step 5: Determine whether the diagnosed ADHD patient belongs to the Attention Deficit Hyperactivity Disorder (ADHD-I), the Hyperactivity-Impulsive Disorder (ADHD-II), or the Mixed Disorder (ADHD-III).

2. The method for diagnosing ADHD and its subtypes based on deep networks according to claim 1, characterized in that, Step one involves building the gaze-tracking network Cognition_Net, specifically through the following steps: Step 1: Embed a multi-scale feature fusion module (EfficientMLCA) and a bidirectional long short-term memory network (Bi-LSTM) into the mid-to-high-level feature outputs of the original backbone network ResNeSt26d, and introduce a wavelet attention feature fusion module (WA) to form the Cognition_Net network. Step 2: Train the Cognition_Net network using the ETH-XGaze dataset.

3. The method for diagnosing ADHD and its subtypes based on deep networks according to claim 1, characterized in that, In step two, attention signals are obtained by inputting a short frontal video of the subject into Cognition_Net. At(t) and iris signal Ir(t) The specific steps are as follows: Step 1: Collect a short frontal video of the person to be examined. I(t) ; Step 2: Input the short video into Cognition_Net to obtain attention signals. At(t) and iris signal Ir(t) .

4. The method for diagnosing ADHD and its subtypes based on deep networks according to claim 1, characterized in that, Statistics in step three At(t) variance ( At_std ), average rate of change ( At_v_mean );statistics Ir(t) average rate of change ( Ir_v_mean The specific steps are as follows: Step 1, Attention signals At(t) Perform Kalman filtering and calculate At(t) variance ( At_std ), average rate of change ( At_v_mean ); Step 2, iris signal Ir(t) Perform wavelet filtering and calculate Ir(t) average rate of change ( Ir_v_mean ).

5. The method for diagnosing ADHD and its subtypes based on deep networks according to claim 1, characterized in that, Step four determines whether the person being tested has ADHD or not, using the following methods: If 0 < At_std<0.14, 0 <At_v_mean<4.05,0<Ir_v_mean<1230, This child is a normal child; otherwise, they would have ADHD.

6. The method for diagnosing ADHD and its subtypes based on deep networks according to claim 1, characterized in that, Step 5: Determine whether the diagnosed ADHD patient belongs to the Attention Deficit Hyperactivity Disorder (ADHD-I), the Hyperactivity-Impulsive Disorder (ADHD-II), or the Mixed Disorder (ADHD-III). The specific steps are as follows: Step 1, if At_std>0.14, 0 <At_v_mean<4.05,Ir_v_mean> 1230, The child has ADHD-I. Step two, if 0 < At_std<0.14, At_v_mean>4.05, Ir_v_mean>1230, The child has ADHD-II. Step 3, if At_std>0.14, At_v_mean>4.05, Ir_v_mean>1230, The child has ADHD-III.

Images