FACIAL PARALYSIS SEVERITY DETECTION USING TRANSFER LEARNING AND ARTIFICAL INTELLIGENCE

Abstract

This project introduces a novel deep learning
based approach to detecting facial paralysis, a
condition that affects millions worldwide but often
goes undiagnosed in its early stages. Leveraging
the InceptionV3 algorithm for its robustness and
efficiency in real-time object detection, we have
developed a model that accurately identifies signs
of facial paralysis from both static images and live
video feeds. The dataset, sourced from Kaggle's
official website, comprises 3,000 images
categorized into two classes, enabling the model to
learn diverse representations of the condition. Our
model achieved an exceptional accuracy of 95.3%
during testing, underscoring its potential as a
reliable diagnostic tool.
The model was trained and validated using Google
Collab, benefiting from its powerful
computational resources and collaborative
environment. Subsequent deployment in a web
application provides an accessible platform for
users to register and obtain authorization from an
administrator. Upon login, users can upload
images or access a live camera feed to detect facial
paralysis, facilitating early diagnosis and
intervention. The web application also features an
administrative module responsible for user
authentication, model training, and performance
analysis, ensuring the system's integrity and
effectiveness.
This project not only showcases the application of
deep learning in addressing critical health issues
but also sets a precedent for the development of
accessible diagnostic tools that can be deployed at
scale, offering significant implications for
telemedicine and remote healthcare services.