The field of diagnostic imaging is constantly advancing and improving with the development of new and innovative technologies being developed. Over the past number of years, the introduction of Artificial Intelligence (AI) has taken over the industry across many modalities. Artificial intelligence (AI) refers to the ability of a machine to simulate human intelligence by thinking and acting like humans. Deep learning is a sub-discipline of AI, which specifically addresses various tasks through building deep neural networks. Deep learning-based AI techniques have been actively investigated in medical imaging, and its potential applications range from data acquisition and image reconstruction to image analysis and understanding.(Wang et al., 2021). This new technology is allowing faster, more accurate diagnosis of diseases than ever before. In an article released by Oren et al. (2020), AI often detects minor image alterations, more relevant outcome variables include new diagnosis of advanced disease, disease requiring treatment, or conditions likely to affect long-term survival. The occurrence of clinically meaningful events—symptoms, need for disease-modifying therapy, and mortality—strongly affect quality of life and should be the focus of AI-based investigations.
One example currently being used in the department I work in, is the use of AI technology to instantly reconstruct images and diagnose a patient experiencing stroke like symptoms. Before the purchase of this software/cloud based server system (RAPIDAI), this required a Technologist to manually scroll through images and highlight different vessels in the CT images that appeared to be of optimal image quality and then proceed to reconstruct the images. The images would then need to be sent to a Radiologist for them to read the images; then a call would be made to the lead neurologist to share the imaging report. The entire process could take at least 30 minutes to complete. As discussed in a previous blog post (for Assignment #2), I mentioned how important it is for quick analysis of information for optimal treatment and recovery for patients who suffer strokes. The new software program allows for images to be sent directly from the CT scanner to the cloud server, where the images are automatically reconstructed. It then creates images which highlight areas of concern (ie: areas of the blood receiving low blood flow – which could signify a stroke) and immediately sends them anyone on the distribution list. The program is set up so that the Radiologist and lead Neurologist get the images via e-mail or on a mobile application on their cell phone. The last modifications we made to the protocol allowed for the team to get images in a long as 5 minutes (the goal is that it’s received in 2 minutes)! Although the physicians still have the ability to review the images in more detail, this software application has lead to instantaneous diagnosis, which allows the team to decide what the next steps of treatment should be. We are soon hoping to have this same software program assist with cardiac events and pulmonary embolisms – but there is still a lot of testing that needs to be done before then.
Below is video explaining how this program works and other information regarding how this program reduces delays in care:
Another benefit of deep learning is related to the reduction of radiation dose to patients. For modalities which obtain images with the use of x-rays (such as CT and PET), the radiation dose delivered to the patient must be limited, as radiation can be harmful to the patient and an excessive dose may lead to the result of secondary cancer. Although it would be easy to suggest that a lower radiation dose is applied, lower radiation doses result in suboptimal image quality, and it may affect the diagnosis accuracy. The benefit of having deep learning is that the network would be able to be “trained” to map between the low dose CT image and the corresponding standard-dose CT image. Once the network is trained, the image quality can be significantly improved by passing the low dose CT image through the network. One algorithm that has been developed has allowed for a 80% reduction in radiation dose, while the detectability of CT image can be improved up to 157%. (Wang et al., 2021).
The important thing to note is that, the more image sets that are loaded into the software system, the better the system gets. With that being said, many facilities who choose to purchase AI software systems do so, but don’t actually utilize them for many years. Images are collected and compared with a Radiologist’s report to start building a database to help identify diseases. Like any software there is often “factory setting” baseline images that these systems have built in, but by building a robust library of examples based on patients’ different body habitus, spread of disease, and appearance on images, the more accurate the AI system is at identifying cancers, deformities, and other illnesses.
One way in which AI helps minimize discrimination and marginalization is that it is a computer based software system which compares the clinical indication entered by the ordering physician with the images taken. There is no bias occurring in the diagnosis since there is no person involved in the process. As mentioned before, the software system actually benefits from patients presenting with difference races, body sizes, and disease progression as it fine tunes itself based on this.
It’s also important to understand that AI is unlikely to replace radiologists, but a radiologist who uses AI might be more productive than a radiologist who does not. (Oren et al., 2020). Although this technology is extremely useful, there will always be a need for Radiologists to ultimately provide the final report for diagnostic imaging procedures.
All in all, there have been significant benefits to having AI in diagnostic imaging as it has shown to be quicker and more accurate in disease diagnosis. Given this technology is still quite new to the industry, the current use in medical imaging are just scratching the surface of all the potential the benefits AI has in store.
More information on AI from the 2021 RSNA (Radiological Society of North America):
Resources:
iSchemaView, I. (n.d.). Stroke patient care: Rapid stroke. Pushing Boundaries in Stroke Patient Care. Retrieved March 24, 2022, from https://www.rapidai.com/stroke
Oren, O., Gersh, B. J., & Bhatt, D. L. (2020, August 24). Artificial Intelligence in medical imaging: Switching from radiographic pathological data to clinically meaningful endpoints. The Lancet Digital Health. Retrieved March 24, 2022, from https://www.sciencedirect.com/science/article/pii/S2589750020301606
Parekh, S. (2021, December 8). Video: Artificial intelligence trends in medical imaging. Artificial Intelligence Trends in Medical Imaging. Retrieved March 23, 2022, from https://www.itnonline.com/videos/video-artificial-intelligence-trends-medical-imaging
Wang, S., Cao, G., Wang, Y., Liao, S., Wang, Q., Shi, J., Li, C., & Shen, D. (2021, December 13). Review and prospect: Artificial intelligence in advanced medical imaging. Frontiers in Radiology. Retrieved March 24, 2022, from https://www.frontiersin.org/articles/10.3389/fradi.2021.781868/full
Comments