How Artificial Intelligence Can Improve CT Scans

Thanks to a 3D camera and artificial intelligence (AI), patients can be positioned optimally for a CT scan. The result: improved image quality and the right dose.

A new patient positioning solution by Siemens Healthineers is improving Computed Tomography (CT) imaging: The "FAST Integrated Workflow" consists of a 3D camera, touch panels and intelligent software. Based on the protocol, the program calculates how the patient needs to be positioned. The patient table then moves precisely to the correct position for scanning. This procedure had previously been time consuming and prone to error; now it can be automated and is less prone to errors. The result is high quality, accurate imaging of the organ to be examined.

95 Percent of all Patients not positioned correctly


Studies show that 95 percent of all patients are not correctly positioned for the CT scan. For a CT scan, the patient must be lying in the exact center of the scanner (isocenter) to ensure best image quality with the lowest dose possible. According to the studies, deviations of as little as a few centimeters can increase image noise or radiation dose. So far, clinical staff has been working with laser markers that are projected onto the body. However, it is often difficult to configure the correct area due to the complex anatomy of the patient. The staff members themselves are also different in size, so that their perspective of the patient and table can lead to deviations, while professional experience also plays a role. “For many years now, our clinical users have been requesting that the scanner be fitted, so to speak, with ‘eyes’, to support patient positioning," explains Thomas Böttger, product manager for scan automation. This would mean that all patients, regardless of whether they are tall, small, sturdy or slim, can with a simple press of a button, be positioned in the scanner directly in the isocenter. Siemens Healthineers has innovated and originated this method, offering the first commercially available solution for automated selection of optimal patient positioning in CT.

A good Combination: FAST 3D Camera and Artificial Intelligence


When the specialists began to look for a solution a few years ago, the combination of images from the FAST 3D Camera whose data is evaluated by AI-based (artificial intelligence based) algorithms proved to be very promising. The imaging specialists at Siemens Healthineers in Princeton, USA, chose the method of Deep Learning for this purpose. Deep Learning, a sub-discipline of AI, uses artificial neural networks to learn from large quantities of data and repeatedly link what they have learned with new content. For the "eyes" of CT scanners, such algorithms have learned, using a large amount of clinical data, to three-dimensionally model the position and angle of the patient on the CT table. The algorithms were developed by Terrence Chen, head of research at Vision Technologies Solutions, and his team. 


Sturdy or Slim – each Body Area is automatically recognized correctly


The FAST 3D Camera, which is mounted on the ceiling above the patient table, takes a picture of the patient already lying down. The software recognizes the body contours of the patient three-dimensionally, and with the aid of an infrared camera, can do so even if the patient is covered i.e. with a blanket or clothes. The system then selects the ideal iso-center position for the selected protocol.


Fine tuning can then manually be performed by moving the marked boundaries with the fingers, although with some intelligent processes, this is rarely required. Now all it needs is the push of a button and the table automatically moves the patient to the optimal position, where scanning is can be carried out.

The software recognizes the body contours of the patient three-dimensionally, and can do so even if the patient is covered.

Large Volumes of Training Data required


“One of the biggest challenges involved is to collect enough training data so that the algorithm learns to recognize patients correctly, regardless of the specific examination situation,” notes Chen. With the help of the clinical collaborations that Siemens Healthineers maintains worldwide, the researchers collected a sufficient amount of data from the recordings of the widest variety of body types and examination situations.


FAST – the acronym stands for "Fully Assisting Scanner Technologies” – is already approved and in clinical use. According to the experts, it will gain in importance: “The number of CT examinations is rising steadily without clinical staff being increased accordingly," explains product manager Böttger. AI should therefore not only increase image quality for a reliable diagnosis, while at the same time keeping radiation exposure as low as possible, but also increase efficiency and help shorten waiting times for the patients.


Katrin Nikolaus

Subscribe to our Newsletter

Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.