The application of AI in the whole process of medical imaging equipment

AI is no longer a commercial gimmick. It has applications in various industries. Today, let's take a look at how AI can accompany and help the development of imaging equipment in the medical industry.

AI mainly solves three problems. First, AI can be used as an experienced doctor to improve the diagnosis and treatment level of grassroots hospitals. Second, AI can be used as an assistant to relieve doctors from complicated work. Third,Explore patterns in big data.

Let's reveal the role of AI from the entire workflow of imaging equipment:

(1) Scanning Stage

Smart Positioning:

The positioning work is mainly completed by the operating technician, who completes the preliminary positioning according to the scanning protocol. This method has low operation efficiency and poses a radiation safety hazard to the scanning technician. In view of this, AI can be used to complete the automatic positioning process.With the help of the high-definition camera installed above the device, the patient's photos are taken, the anatomical key points are automatically identified, and the bed movement size is automatically calculated according to the camera calibration parameters and scanning protocol.

Automatically position images:

The positioning process is mainly based on the low-dose positioning image to accurately set the scanning range and angle. Compared with CT, MR can scan at any angle, so the angle information is very important. For example, by dividing the headSagittal image of the corpus callosum, with the rotation of the bounding box angle through the corpus callosum as the scan angle.

Image reconstruction:

After the scanning is completed, the CT detector receives the X-ray attenuation signal after passing through the human body, and the MR receiving coil receives the tissue hydrogen proton relaxation and releases the energy signal, which is converted into digital signal by analog-to-digital conversion. The reconstruction process is to convert these digital signals into humanAn understandable grayscale image.

CT image quality is related to radiation dose, and MR image quality is related to scan time. The reconstruction algorithm is to solve how to obtain high-quality images at low dose and in a short time.

Reconstruction algorithms can be divided into: algebraic reconstruction, filtered back-projection, iterative reconstruction, and deep learning reconstruction according to their development history.

There are also two types of reconstruction algorithms based on deep learning. One uses raw raw data as input and outputs high-quality reconstructed grayscale images. Typical examples are GE's Turefidelity and Canon's Aice.

Turefidelity is taken as an example. Its training data uses two scanning parameters to obtain high-quality and low-quality images respectively. The model takes the low-quality image as input, and compares the predicted output image with the scanned high-quality image to calculate the loss function to optimize the parameters..

The other is the reconstructed grayscaleThe image is used as input to perform operations such as denoising, de-artifacting, super-resolution, etc.

(2) Stage of diagnosis and treatment:

There are countless applications of AI in the diagnosis and treatment stage. According to the application in the field of vision, it can be divided into image classification, target detection, image segmentation, image generation, etc.

Image classification: As a qualitative analysis method, it can provide doctors with a basis for diagnosis, diagnosis of benign and malignant tumors, prediction of bone age, rapid classification of stroke, etc.

Object detection: As a semi-quantitative analysis method, it can not only provide classification results, but also provide the location and size of ROI, such as lung nodule detection.

Image segmentation: As a full quantitative analysis method, it can provide pixel-level classification results, and then calculate ROI volume, maximum diameter and other parameters, such as tumor segmentation, organ segmentation, etc.

Image generation: It mainly solves the problem of insufficient medical image data.