- Nikhil K J
- Femina N
- Chithira M
- Arun J Dev
- ASD and ADHD are two common neurodevelopmental disorders that affect millions of individuals worldwide.
- Deep learning improves medical diagnosis accuracy and efficiency.
- Deep learning algorithms can identify patterns in clinical data which can be trained on large datasets and can predict neurodevelopmental disorders.
- This approach has the potential to enhance the accuracy and reliability of diagnosis, which is critical for early intervention and improving outcomes for individuals with these conditions.
- ASD diagnosis involves a comprehensive assessment of an individual's behavioral, social, and developmental functioning.
- ADHD diagnosis involves a comprehensive assessment of an individual's behavioral, cognitive, and developmental functioning.
- Current diagnostic practices rely on interviews, observation, and rating scales that can be time-consuming and require trained professionals.
- The readings of EEG signals are analyzed by neurologists to detect and categorize the patterns of the disorder.
- The visual examination is time-consuming and laborious and it requires the services of an expert.
- Absence of an implemented automated system to detect the neurological disorders at an early stage of life.
- Diagnosis can be complicated by overlapping symptoms between ASD and ADHD, as well as comorbid conditions.
- To select an alternative method for the detection of neurological disorder by performing classification using EEG signals based on deep learning to get a relatively high degree of accuracy.
- Our work mainly analyzes the performance of various deep convolutional architectures for the classification of ASD and ADHD.
- Many lives are impacted by undetected neurological disorders in their early stages. Our project aims to change this through an innovative AI-based classification system.
- Noise and artifacts can reduce EEG signal quality and impact classification.
- Makoto's pre-processing pipeline and the EEGLAB toolbox in MATLAB will be used to address this.
- Pre-processing improves data quality, reduces noise, and saves time for more accurate analysis.
- Apply a notch filter to remove unwanted frequencies from the data.
- Apply a bandpass filter with a range of 0.3-50Hz to isolate frequencies of interest for the analysis.
- Re-reference the data by converting the reference to average reference.
- MNE-Python is a Python-based software package for analyzing EEG and MEG data.
- It includes functions for loading, preprocessing, visualizing, and analyzing neurophysiological data.
- Artifact detection and correction functions in MNE-Python can identify and remove eye blinks, muscle activity, and other unwanted noise.
- Epoching functions can segment the data into smaller time windows for further analysis.
- As much as more training samples be available, machine learning models can better estimate the relationships between features.
- EEG data is typically collected for several minutes at a time for a single participant.
- By splitting the data into shorter segments, better control of the complexity of the input is possible.
- It is much easier for the deep learning model to learn the relevant patterns in the data.
- Will use Hanning Window to split the samples into 1-second segments with 50% overlap, which increases the number of samples for the proposed method.
- Data segmentation can be implemented using tools such as MNE-Python and EEGLAB.
- The datasets received from ICCONS Shornur include 16 pure ASD and pure ADHD signals of children between the age group of 2-10 years.
- Each signal consists of 21 channels and is sampled at 250Hz.
- The dataset consisted of EEG recordings from 16 patients with ASD (Autism Spectrum Disorder) and 16 patients with ADHD (Attention-Deficit/Hyperactivity Disorder).
- Two patients with ASD and two patients with ADHD were randomly selected from the dataset to be part of the testing dataset.
- The remaining 14 patients with ASD and 14 patients with ADHD were used to train and validate the deep learning model.
- The training dataset trains the model to classify ASD and ADHD, and the testing dataset evaluates the model's performance on new data.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent