This repo contains the computational trials of the research paper titled: Few-Shot Learning for Early Detection of Autism Spectrum Disorder in Children
Autism Spectrum Disorder (ASD) is a developmental condition affecting communication, behavior, and social interactions, characterized by a range of symptoms and severity levels where early detection and intervention are crucial for effective support. Acquiring labeled data is challenging due to high costs, specialized knowledge, or sample scarcity, especially in rare conditions
like childhood ASD
.
We utilized few-shot learning (FSL) which is a machine-learning technique that trains the models by meta-learning with a minimal amount of labeled data, often just a few examples for each class.
- Few-shot learning for handling limited labeled data.
- Pre-trained backbone model for feature extraction.
- Minimizing processing Time.
- High accuracy in autism detection from children's images.
pip install -r requirements.txt
git clone https://github.com/Arwa-Fawzy/Few-Shot-Learning-for-Early-Diagnosis-of-Autism-Spectrum-Disorder-in-Children.git
cd Few-Shot-Learning-for-Early-Diagnosis-of-Autism-Spectrum-Disorder-in-Children
In our FSL model, a labeled support set
S = {(x_i, y_i)}_i=0^(N*K) comprises approximately 3000 images from a public kaggle dataset Autistic Children Facial Data Set. , simulating a catalog or primary training data, alongside a query set
Q = {(x_i, y_i)} for i = 0 to M resembling testing data, and 'episodes or tasks' that are equivalent to epochs. Each image in the query set requires classification using labels provided in the support set. Figure Shown illustrates a 2-way, 4-shots
with 10 query
instances of the classification task, where '2-way' denotes two distinct classes (autistic and non-autistic) and '4-shots' signifies four instances per class for each episodic training.
Our future work will extend this approach to video ASD datasets. Given that autism is better diagnosed through long-term observation rather than static images, we aim to develop an FSL model capable of analyzing video data. This approach will utilize temporal patterns and behavioral cues observed over extended periods, which may enhance the reliability of early ASD diagnosis.
Submitted to The 6th Novel Intelligent and Leading Emerging Sciences Conference IEEE