Introduction

Welcome to the module CNN – Industry Applications. Let’s first talk about what you are going to learn in the upcoming sessions.

Objectives of this Session

Here is a quick run- through of the broad learning objectives of this session:

1. You will get a hands-on experience in building an end-to-end pipeline for training CNNs. This is almost exactly how you would do this in a production environment. You are provided code throughout this session. We urge you to try experimenting with this code – it’s the best way to make yourselves proficient in image processing techniques.
2. We will spend a good amount of time on data preprocessing techniques commonly used with image processing. This is because preprocessing takes about 50-80% of your time in most deep learning projects, and knowing some useful tricks will help you a lot in your projects. We will first use the flowers dataset from Kaggle to demonstrate the key concepts. Later, we will apply the same techniques on Chest X-ray images. The purpose of starting with the flowers dataset is to understand the process using images that you understand before getting into medical images.
3. Eventually, we will build a classifier – for both the Flowers and X-ray datasets. We’ll take you through the important steps and hyperparameters involved in this process.

Neural Networks in Industry Applications

Neural Networks have changed the face of image processing in the industry. Through this demonstration, we’ll see how they are used in the medical imaging industry.

Some of the notable types of medical images are:

1. X-rays
2. CT Scans
3. MRI images

Before we go any further, let’s get to know our industry expert. Today, jobs related to image processing are specialised enough that deep learning experts need to also understand the domain where machine learning is being applied. Rohit has worked extensively with medical images and will demonstrate how that knowledge can be applied to deep learning tasks (using chest X-Rays as examples).

Structure of this Session

This session is divided into two parts:

• Flower classification: First, we will see how to classify flowers into “roses” and “daisies”. This is a toy dataset and its purpose is to introduce you to the key concepts and methodologies. In this session, you will learn:
• How to set-up an end-to-end pipeline for training deep learning models.
• Preprocessing techniques: Morphological transformations etc.
• Data augmentation using data generators
• Building a network: Ablation experiments, hyperparameter tuning, storing the best model in disk etc.
• X-ray classification: We will apply the concepts learnt in the first half to Chest X-ray images. Here, you will learn how to identify and debug problems often encountered during training.

People You Will Hear From In This Session

Rohit Ghosh

AI Researcher, Qure. ai

Additional Reading

1. If the history of science piques your interest, here is a short article on the history of the first X-ray image.
2. Wondered what the difference between a CT Scan and an MRI scan is? Here is an easy answer in layperson language.

Report an error