We have a university project involving 163 DICOM breast images, some of which contain benign tumors and others malignant ones. We want to implement a machine learning model using CNN that, upon seeing each image, can tell whether the tumor is benign or malignant. We must implement this code in two separate files: one where we build a CNN model from scratch, and another where we use a pre-trained CNN model. Then we compare the two models to see which performs better. I want to code this project in Python using Google Colab. I plan to upload the images from my Google Drive. The benign or malignant labels are specified in the metadata. All images are in .dcm format and must be resized to 224×224 pixels. For the pre-trained model, we have to use ResNet-50. We must use 80% of the data for training and 20% for testing. Normalization from 0 to 1 must be applied. All images (benign and malignant) are located in a single folder. Specificity and sensitivity are important metrics for us. Would you please help me? I am so beginner in Machine learning. I have no idea how to start it. Continue reading...
