Deep Learning Models for Mammograms

Dataset

MIAS

Download the dataset from this link

Image Preprocess

1. Convert image from pgm format to png

   There are multiple ways to convert images to png. I preferred to use mogrify from ImageMagick in command line

   mogrify -format png path/to/files/*.pgm

2. Remove Noise

   In order to remove the noise from the image the biggest contour(largest blob) is found from the image. Mask of the biggest contour is used to clean the images

   Code

Data Preparation

Images are split into benign, malignant and normal. Train and test sets are split at 80:20

Code

Models

VGG16

Vgg16 is used to run the classifcation model. The model is finetuned to classify malignant, benign and normal images

Code