The idea of a doctor peering into the human body without making a single incision, which once seemed miraculous, is now a reality, thanks to artificial intelligence-driven technologies. Radiologists today heavily use AI-driven image analysis techniques to automatically recognize differences in body scans that are even undetectable to the human eye. X-rays, ultrasounds, PET, MRI, and CT scans, can all be autonomously processed using appropriately trained algorithms, to make initial observations and first-line diagnoses. For example, image segmentation and registration algorithms can extract regions of interest from scans, align and establish spatial and temporal correspondence between different images and modalities, and assist doctors in making a quick and accurate diagnosis of clinical conditions. What was once a time-consuming, laborious, and error-prone process, is now more robust, efficient, and accurate.
Artificial Intelligence can contribute to the field of medical imaging in two ways:
- Computer-aided Diagnostics – this refers to the use of trained, image analysis algorithms to detect and quantify radiographic anomalies. Computer-aided diagnostic systems assist doctors in making quick and accurate diagnoses of complex conditions. For example, convolutional neural networks, an AI technique, is shown to facilitate the early detection of neurological abnormalities in patients afflicted with Alzheimer’s disease .
Figure 1: Computer-aided diagnosis
Robot-assisted Medical Imaging – this refers to the use of a robot to acquire a medical image. Some examples are robot-assisted ultrasound imaging where the transducer is positioned by a robot and robotic capsule endoscopy where the capsule is navigated robotically through electromagnetic control. Robotic medical imaging systems offer higher operator precision and control. This can largely reduce procedural time and radiation exposure.
Figure 2: Robot-assisted medical imaging
To summarise, here are some benefits that AI can offer in the area of medical imaging.
1. Enhanced Productivity via Automation
AI enables automation or semi-automation of parts of the radiology workflow such as treatment planning, dose tracking, radiation estimation, reporting, image sorting, analysis, and diagnostics. This largely reduces repetitive manual work, improves speed and accuracy, and enhances diagnostic capacity, leading to more effective patient care.
2. Improved Diagnosis and Treatment
AI may not completely replace physicians and medical experts, but can greatly assist them in diagnosis and treatment. AI algorithms can be used to detect and flag radiographic abnormalities, triage critical findings and make recommendations based on analysis of large repositories of historical imaging data. Recent studies have also shown that AI systems can accurately detect and diagnose certain types of cancer, cardiac conditions, neurological diseases, thoracic complications, and fractures . Automation of parts of the diagnostic workflow eases the workload on physicians and enables quicker decision-making with fewer errors. AI also supports physicians in building effective treatment plans by enabling automatic risk assessment, dose tracking and optimization, and recommendations based on comparisons with similar cases.
3. Informed Second Opinion for Complicated Cases
AI algorithms can provide well-informed second opinions on problematic medical images, thereby assisting radiologists in decision-making for complicated cases. For example, AI algorithms have been shown to give useful second opinions on stroke patients in hospitals in the UK . X-rays of stroke patients are scanned using AI to identify those with blood clots that require immediate surgical removal. This largely helps in saving patient lives and avoiding serious disabilities.
4. Seamless Clinical decision support
Radiologists and clinicians should be able to communicate seamlessly for effective patient care. Artificial Intelligence can help enhance existing clinical decision support systems for better patient-centric, value-based imaging. AI-driven, semi-automated, radiology workflows can help in quick and smooth communication between physicians and radiologists and promote collaborative working and in-tandem diagnosis.
5. Reduced inter and intra-observer variability
Inter-observer variability is the difference in measurements between observers and intra-observer variability refers to the difference in repeated measurements by the same observer. Radiologists are often hindered in their diagnoses by both of these, as human perception lends itself to some form of variability. Automation of medical image processing using AI-driven algorithms reduces, and to a large extent, even eliminates inter-observer variability and improves the reproducibility of measurements as shown in studies. This enables radiologists to concur with each other and make accurate diagnoses.
Challenges of AI in Medical Imaging
We have seen the benefits of using AI in medical imaging. However, leveraging AI effectively in medical imaging comes with its own challenges. We describe some of these below.
- Dealing with 3D reality – Deep-learning models are currently trained on 2D images. CT and MRI images are usually in 3D, adding another dimension to the problem. Due to their projected nature, most deep learning algorithms are not adjusted to conventional X-ray images.
- Non-standardized image acquisition — It is a challenge to train AI algorithms on medical images acquired with different scanner types, acquisition settings, and non-standardized acquisition methods.
- A smooth user experience — Current radiology software is generally not very user-friendly. Once we are in the program, we can’t live without the manual since it requires too many clicks. If AI companies want their software to be used in clinics, they must create user-friendly software.
The future: what does it look like?
Medical imaging research clearly demonstrates the effectiveness of artificial intelligence as an essential tool for improved patient care. AI facilitates the automated and accurate detection of abnormalities in the body and the formulation of effective treatment plans. Instead of feeling pushed out by machine intelligence, radiologists should learn to engage with and promote AI as a tool that complements their expertise.
In recent years, several companies have transitioned the use of AI in medical imaging research into industry opportunities through the development of AI-driven advanced medical imaging tools and products. The radiology field is expected to undergo more such rapid advancements and the future looks promising with AI enabling patient-centric, value-based medical imaging and diagnostics.
 Researchers Leverage AI to Detect Causes of Alzheimer’s Disease, Acta Neuropathologica Communications, Mount Sinai Health System, September 2022, https://healthitanalytics.com/news/researchers-leverage-ai-to-detect-causes-of-alzheimers-disease
 Artificial Intelligence Bolsters Colorectal Cancer Detection, https://healthitanalytics.com/news/artificial-intelligence-bolsters-colorectal-cancer-detection
 AI gives Doctors Second Opinion on Stroke Cases, https://www.thetimes.co.uk/article/ai-gives-doctors-second-opinion-on-stroke-cases-z2vzxnp02
 Effect of AI-assisted software on inter- and intra-observer variability for the X-ray bone age assessment of preschool children, https://pubmed.ncbi.nlm.nih.gov/36348326/
Edited By : Naga Vydyanathan