Technology

Data Science

Industry

Healthcare

Facial Palsy

Image recognition and neural network training helps in detecting Facial Palsy. It is a weakness occurring on the facial muscles where there is loss of facial movement due to the occurrence of nerve damage. The affected areas can be categorized into three where it can be either lower half or one side of the face or in some cases, it could be a whole face. The point annotations are marked upon the faces and the important points assist in determining the palsy with the help of annotation tools, neural networks and Distance analysis.

Client Profile

Our client is a leading healthcare organization in the field of Artificial Intelligence. They assist the patients by providing state-of-the-art services on creating efficient software to facilitate the doctors and the patients. And Detection of Facial Palsy is one among them.

Objective

The main objective is to train the neural networks for classifying the normal faces with the palsy ones. It diagnoses the palsy in an early stage to avoid the criticality of a patient.

Key Challenges

The major challenge faced was when a face smiles, it reacts similarly to facial palsy. And so, the datasets had to be provided very precisely for training. Additionally, the image recognition process needs to be very accurate. A huge collection of input images with palsy had to be collected for accurate detection.

Our Approach

We approach in a manner to detect the image and find the earlier palsy stages by the recognition of several structural changes in the points of apex and their distance on either side of the face. The nasal bridge and the tip of the nose becomes the apex for the face to be examined and the relevant points are plotted either side of the face. The distance between the apex and respective points are vectorised and the result is studied for symmetry.

Our Solution

The input dataset is provided by employing annotated points and their distance between them. The annotated images are then marked into a neural network (RCNN) and to be trained. The classes with different categories are labelled onto the neural networks for maximizing the accuracy.

Technology Used

  • Label IMG (Annotation Tool)
  • Neural networks
  • RCNN
  • Fast RCNN
  • Faster RCNN
  • Darknet
  • Distance Analysis

Results

Early-stage detection of Facial palsy acts as a boon to the patients and the doctors, avoiding their entry into critical stages.