Classification of Medical Images based on Texture Analysis

MM61503-Digital Image Processing and Applications Autumn 2021 Semester – End-Term Evaluation

• The Dataset used for the classification- Medical MNIST.
• Classification of Medical Images based on Texture Analysis is implemented based on the gray-level co-occurrence matrix and its properties. Six major classes of images (Abdomen CT, Breast MRI, Chest CT, Chest X-Ray, Hand X-Ray and Head CT) are classified in this implementation.
• Batch extraction of the images in the dataset is done using the separately written function in MATLAB which can be found here- Medical MNIST Dataset-Batch Feature Extraction.
• The properties of Energy/ Uniformity which gives the sum of squared elements in the GLCM is extracted from the gray-level co-occurrence matrix and additionally Entropy (statistical measure of randomness used to characterize the texture of the image) is used to classify the image type as these were shown to provide the best distinction among different class of images via statistical analysis.
• Aforementioned data from 54,000 images are extracted to give the range for classification.
• The function receives the image variable as the input and returns the type of image as output.

The following steps are implemented within the function

1. Checking if the input image is in gray format.
2. Creating the gray-level co-occurrence matrix of the input image.
3. Extracting the properties of the gray-level co-occurrence matrix into a cell array.
4. Using the obtained values from the image to compare with the preset parameters obtained by extracting data from 54,000 test images and returning the Image type.