Problem Statemen

The evolution of artificial intelligence is great, and it is increasingly impacting the way of working in different areas. Such is the case of the health area, where it is revolutionizing care in areas such as:

• Diagnosis of diseases with medical images.
• Surgical robots.
• Maximizing hospital efficiency.

The Artificial Intelligence healthcare market is expected to reach $45.2 billion by 2026.

The current valuation is $4.9 billion.

The application of Artificial Intelligence, mainly deep learning, has been shown to be superior in detecting diseases from X-rays, MRIs and CT scans, which could significantly improve the speed and accuracy of diagnosis.

For a better approach: https://research.aimultiple.com/looking-for-better-medical-imaging-for-early-diagnostic-and-monitoring-contact-the-leading-vendors-here/

In this case study we will assume that we are working as an AI/ML consultant and have been hired by a medical diagnostic company in New York.

It has been assigned the task of improving the speed and accuracy of brain tumor detection and localization based on magnetic resonance imaging.

The goal is to reduce cost and aid in early diagnosis of tumors, which would essentially increase the likelihood of successful treatment.

The team was given the task of collecting magnetic resonance images of the brain, a total of 3,929 brain images with which they have to develop the algorithm.

Original data source: https://www.kaggle.com/datasets/mateuszbuda/lgg-mri-segmentation

Deep learning pipeline to perform classification and localization

Two stages will be carried out to detect and locate the brain tumor (if there is one).

1. Use of a ResNet to classify images into:

   • 0: There is no tumor.
   • 1: There is a tumor.

2. Once this process is finished, the next stage is different depending on the result obtained in the first stage:

   • If the result is 1: The image will serve as input for a segmentation model supported by a ResUNet to help us locate the tumor.
   • If the result is 0: It is considered as a healthy patient, so it is not necessary to submit the image to the second stage.

A review of, what is image segmentation

The goal of image segmentation is to understand and extract information from images at the pixel level. It can be used for object recognition and location, offering great value in applications related to medical imaging and self-driving cars.

The objective of image segmentation is to train a neural network to produce a pixel mask of the image, highlighting the part where the object to be detected and located is located.

Note:

Click on the links below to access the complete project. We put at your disposal an English and a Spanish version, both to be downloaded in pdf. In addition, you have access to the repository where you will find the respective Notebooks.