This project is aimed at developing two deep learning models for each of the classification problem, one model would be based on any existing deep neural network and the other model would be our own deep neural network. A web application and all the models were developed in Python using mainly the following important packages:
- Keras for deep learning models
- Streamlit for web application
- Numpy, Scikit-learn, Matplotlib – provide visualisation and other utilities The implemented models are then compared using evaluation metrics like Accuracy, precision, recall, F1 score, ROC AUC score, Confusion matrix etc.
The dataset for this project is BloodMNIST and BreastMNIST which are loaded from MedMNIST2D dataset.
- BloodMNIST consists of microscopic images of blood samples collected from healthy individuals without any infection, hematologic or oncologic diseases and not undergoing any pharmacological treatments at the time of blood collection.
- BreastMNIST is a collection of ultrasound images of breasts for identifying any abnormalities. It is originally categorised as normal, benign and malignant but due to low resolution images it is simplified into a binary problem by combining normal and benign as positive and malignant as negative.