Before the use of AI, Radiation therapists must ensure that all risk and target structures relevant to treatment planning have been appropriately marked.
But now, the tedious drawing of hundreds of contours on the layers of a computed tomography can be easily automated with deep learning, says chaktty.
Radiation therapy is particularly suitable to show where the new technologies can already make valuable contributions today.
What is Deep Learning?
Deep learning, according to Techpally Magazine, is the term used to describe modern machine learning methods that can model highly complex relationships between raw data such as texts and images and their assessment by medical experts.
The fact that cognitive tasks are solved with the computer means that this research area automatically falls under “artificial intelligence”, although the methods developed by no means meet the claim to be intelligent.
The particularly successful artificial neural networks that are currently contributing to revolutionary developments in computer science can be understood as particularly complex statistical models.
The term “learning” can also easily lead to misunderstandings: It is more of a consciously externally controlled training.
In which specific medical tasks can neural networks support?
In principle, almost any data can be processed, such as texts, laboratory values, medical image data, right up to the complete digitized patient history.
In practice, however, with the complexity of the task, the required amount of training data quickly increases beyond realistic dimensions, says Techpally.
Deep learning is known to require millions of training examples, but this magnitude does not seem attainable in medicine at first glance.
However, if you look at the drawings of objects in medical image data, you ultimately have to decide for each pixel whether it belongs to an organ, a vessel, or a tumor, for example. So every single point becomes a training example for the contouring task.
AI is Very promising for the future
Deep learning promises much more than just automated contouring in the future, but many important questions are still open research topics, according to Healthpally boss when interviewed.
For example, if the computer is to suggest complex diagnoses that include not just individual images but the complete patient history, the traceability and explainability of the results is an important prerequisite for practical use.
Here, too, contouring is a possible intermediate step, the results of which are used for quantitative analyzes.
If necessary, the structures can easily be checked by humans for correction and perfection.
Already today, instead of having to laboriously draw hundreds of contours for treatment planning manually, radio-oncology specialists can rely on better computer support from neural networks without the computer having to develop real intelligence or the human being having to give up responsibility.
In the future, patients will also benefit from this when a more rapid adaptation of the plans for radiation therapy enables individual optimization of the individual sessions.
Artificial intelligence has influenced almost every sector, in health, the impacts is very great after technology.
Radiation therapy is one of the verticals of health where deep learning and neural network has impacted, and drawings can now be automated and in the future, the whole processed would be controlled by artificial intelligence.