AI in radiology: top 10 use cases & best practices

AI in radiology: top 10 use cases & best practices

March 22, 2024

Within the past decade, rapid technological advancements have happened in both medical imaging and artificial intelligence, causing the two to converge in recent years. Major advances in computing power coupled with enhanced access to image data has accelerated AI research in different fields of radiology. In this article, we dive deep into the applications of AI in radiology, discuss its benefits, and provide a market overview.

 

What is AI in radiology?
Radiology is a medical specialty that uses imaging technology to diagnose and treat illnesses. Artificial intelligence is a self-learning computer software that can analyze images in different modalities and help quickly spot anomalies. AI-based systems have proven to be more accurate than human radiologists in detecting and diagnosing diseases, which boosted their widespread adoption in radiology departments globally.

According to a recent study, there is a significant workforce shortage of radiologists: 1.9 radiologists per million people in low-income and 97.9 in high-income countries, respectively. UK researchers asked expert clinicians to classify more than 3,600 images of hip fractures. The study showed that clinicians recognized only 77.5% of images accurately, while the machine learning algorithm achieved 92% accuracy. In a nutshell, the ever-growing demand for radiology coupled with increasingly more accurate AI-based radiology systems makes AI a saving grace for healthcare worldwide.

AI in radiology market statistics

of radiologists use AI in their clinical practices

American College of Radiology

plan to purchase AI tools in the near future

American College of Radiology

AI in the global radiology market CAGR from 2021 to 2030

Vision Research Reports

Chart title: Artificial intelligence in radiology: market size forecast, 2021-2030 Data source: Vision Research Reports

10 AI use cases in radiology

The number of use cases of artificial intelligence software in clinical data science and radiology practice is growing. Let’s discuss the top 10 applications of AI in radiology:

Enhancing cardiac imaging

AI has multiple applications when it comes to cardiac health. One such application is predicting which patients could develop irregular heartbeats by analyzing electrocardiograms. Another application of AI lies in cardiac imaging. AI can enhance visualization of the heart by coloring heart chambers on grayscale echocardiography images in real-time, which streamlines radiology workflows. Philips developed an AI-based system, HeartModel, that is able to produce a colored heart model and project its dynamic 3D representation with wall motions and LV and LA volume changes during the heart cycle.

Classifying brain tumors

In order to identify the type of tumor, doctors take a biopsy, MRI scans, and blood tests. Once the type of tumor is determined, they can use radiomics AI tools to further stratify the tumor into one of several grades. AI accurately classified brain tumors into grades with very few false positives or negatives. A study on intraoperative diagnosis showed that AI can classify brain tumors in under 150 seconds compared to 20-30 minutes for conventional methods. This means that doctors can rely on AI as an additional tool to support their clinical decisions about the best treatment.

Spotting vertebral fractures

Spinal fractures are the most common fragility fractures and can be an early sign of osteoporosis. They can go unnoticed since 54% of vertebral fractures remain undetected by radiologists in CT scans. The researchers at UCB Pharma attempted to solve this issue by training a deep-learning algorithm to find vertebral fractures and grade them. The algorithm was trained on images containing 969 vertebrae coming from three different types of scanners. The algorithm produced an area under the curve (AUC) of 0.93, meaning that it can be potentially used in examination rooms.

Image title: Image preparation for vertebrae detection Data source: sciencedirect.com — Using artificial intelligence to diagnose fresh osteoporotic vertebral fractures on magnetic resonance images
Spotting vertebral fractures
(A): Normal, fresh, and old fractured vertebrae were colored white, red, and blue, respectively. (B): Image divided into several squares by sectioning the images at the same height according to the width of the original image for deep learning. Base; The original image,1-3; Three divided images generated from the original image

Detecting Alzheimer’s disease

Another challenging neurological disease to diagnose is Alzheimer’s. To boost radiologists’ confidence in diagnosing, researchers at the University of California have developed an algorithm that can detect Alzheimer's based on a fluorodeoxyglucose (FDG)-positron emission tomography (PET) scan. This algorithm looks for subtle processes and global changes in the brain (such as changes in glucose uptake), which cannot be observed with the naked eye. After training the algorithm, researchers tested its performance on 188 new images. AI correctly identified 92% of patients with Alzheimer's.

A critical point of using computer-aided detection in Alzheimer’s diagnostics is that the algorithm can pick up very early indications of the disease, which radiologists would not be able to detect. If diagnosed early enough, the progression of the disease can be delayed or even stopped.

Detecting Alzheimer’s disease
Image title: FDG-PET imaging of a patient with AD Data source: semanticscholar.org – Brain imaging in Alzheimer disease, 2012

Diagnosing ALS

A degenerative neurological disease can be devastating, but its early detection allows doctors to draw up an effective long-term patient care plan. Amyotrophic lateral sclerosis (ALS) is a fatal degenerative disease that differs radically from PLS (a non-fatal variation). Diagnostics of these diseases rely on image analysis, where radiologists decide whether the existing lesions are actual ALS lesions or just mimicking them, thus pointing them to PLS. False positives are very common in this area. Using sophisticated machine learning models, it becomes possible to identify risk ratios for evidence of ALS or PLS. According to a study published in Frontiers in Neuroscience, several machine learning-based methods can successfully diagnose ALS.

Assisting with radiology reporting & data-related tasks

Reporting is a time-consuming and error-prone task and is therefore often a source of frustration for radiologists. Furthermore, there are no set-in-stone reporting standards, which leads to variability and a lack of compatibility between data submitted by radiologists. Natural language processing tools offer valuable capabilities for streamlining radiology reporting, from a quick transcription of speech into text to automated report compilation and their logical structuring for improved comprehension. In addition to radiology reporting itself, AI-based solutions can perform related tasks, such as enhancing the quality of scans.

Detecting breast cancer

In 2020, there were 7.8 million women diagnosed with breast cancer, making it the most widespread type of cancer globally, according to WHO. Recent studies point out that AI shows increasingly promising results in oncology for detecting breast cancer  in particular. 

For example, German AI startup Vara in collaboration with The Mammography Reference Centre North in Oldenburg recently conducted a study where researchers assessed 2,396 screening mammograms from women who were later diagnosed with interval cancer. AI correctly detected and localized 27.5% of false negatives and 12.2% of the minimal sign cancers. This means that AI can detect breast cancer signs that aren't visible to many radiologists at the earliest stages of the disease.

Dose optimization

Radiation dose optimization is crucial in pediatric radiology, as excessive ionizing radiation has proven harmful to children and can cause cancer. A 2022 systematic review on the AI for radiation dose optimization in pediatric radiology that assessed 16 peer-reviewed studies on the matter found that half of the proposed AI models achieved dose reductions between 36% and 70%, with the potential to go up to 95%. This means AI can significantly reduce the harmful effect of ionizing radiation on children.

Detecting pneumonia

Radiology is a common method for detecting pneumonia, a deadly lung disease. The problem with pneumonia detection using clinical imaging is that it is often hard for medical professionals to distinguish it from other lung diseases like bronchitis. AI systems, on the other hand, can detect and segment areas of opacity or consolidation indicative of pneumonia, identifying it with greater accuracy. The global pandemic has accelerated research of lung-related diseases in general and pneumonia in particular. For example, researchers have developed a CNN-based model that can achieve 98% accuracy for the detection of COVID-19-induced pneumonia.
Image title: Heatmap COVID-19-induced pneumonia vs heatmap regular pneumonia Data source: hindawi.com — Deep learning-aided automated pneumonia detection and classification using CXR scans
COVID-19-induced pneumonia
a) COVID-19-induced pneumonia
regular pneumonia
b) regular pneumonia

Detecting LVO

Large Vessel Occlusion (LVO) strokes happen when a major artery in the brain is blocked, making it one of the most severe kinds of strokes that is often associated with an increased risk of death or long-term disability. AI solutions for image segmentation can process MRA and CT images to identify and isolate the blood vessels in medical images for precise localization and characterization of potential occlusions. AI algorithms can then analyze the morphology, size, and integrity of the blood vessels, helping radiologists to reliably diagnose and triage LVO strokes. AI has already taken over LVO detection with 16 FDA-approved AI-based tools available on the market, with several studies demonstrating its superior accuracy compared to even the most experienced neuroradiologists.

Looking for reliable AI consultants?

Get in touch

Examples of AI in radiology

Viz LVO

Viz, a medical imaging company that is dedicated to optimizing emergency treatment using deep learning technology, developed Viz LVO, a clinically validated AI-based technology for automation of LVO stroke detection and triage. In 2018, Viz LVO was approved by the FDA and consequentially created a new category of medical devices: computer-assisted triage. The device can automatically analyze computer tomography angiography (CTA) of the brain and identify LVO. A peer-reviewed study found that Viz LVO has a high sensitivity of 96.3% and specificity of 93.8%, making it a reliable companion for neuroradiologists.
Image title: The Viz LVO device offers automatic stroke detection and notification (left), coupled with a mobile DICOM viewer (right) Data source: openaccessjournals.com — AI-powered stroke triage system performance in the wild, 2020
The Viz LVO device offers automatic stroke detection and notification (left), coupled with a mobile DICOM viewer (right)

Lunit INSIGHT MMG

Lunit INSIGHT MMG is AI-based software that helps radiologists more accurately assess mammograms. The deep learning solution can detect suspicious lesions in mammography images and distinguish tumor areas by providing the location of the lesion. Lunit software detects breast cancer on mammograms with 96% accuracy. The solution can help radiologists detect more breast cancer cases and faster, improve the speed of triage, and enhance the reading performance.
Lunit in action
Image title: Lunit in action Data source: lunit.io — Product Brochure Lunit INSIGHT MMG

Lunit's AI tool is particularly effective in helping diagnose dense breasts, which are often more likely to be misdiagnosed among East Asian women, and has helped improve communication between radiologists and breast surgeons.

Yeh Wei Cheng

Yeh Wei Cheng

director of the Department of Radiology Nantou Hospital

Rapid ICH

An intracranial hemorrhage (ICH) refers to bleeding between the brain tissue and skull that can cause severe brain damage and death. Rapid ICH, developed by RapidAI, a global leader in AI-led solutions for vascular and neurovascular conditions, allows medical professionals to make fast, evidence-based decisions in critical situations and accurately detect bleeding on non-contrast CT images. Rapid ICH’s AI-based solution can identify hemorrhages with a sensitivity of 95% and specificity of 94% in as few as 3 minutes. Given that any hemorrhage requires a medical emergency, Rapid ICH automatically sends notifications via a dedicated mobile app, alerting clinicians about critical cases.
Suspected hemorrhage
Image title: Suspected hemorrhage Data source: rapidai.com — Rapid ICH and Rapid Hyperdensity

qXR

qXR, developed by Indian AI startup Qure, is a tool that helps radiologists detect chest pathologies indicative of lung cancer in radiography imaging. With the help of a proprietary deep learning model trained on more than 3.5 million X-rays, xQR can reliably identify abnormalities in the lungs, pleura, mediastinum, and bones in under 1 minute for each scan. A recent peer-reviewed study revealed xQR’s sensitivity of 99% and specificity ranged from 87% to 92%, making it a reliable assistant in detecting malignant and non-malignant lung nodules, tuberculosis, coronavirus, and other lung diseases.

AI-assisted chest X-rays
Image title: AI-assisted chest X-rays Data source: qure.ai — AI for Chest X-rays

10 benefits of AI in radiology

Early detection

Improved prioritization

Improved accuracy

Optimized radiology dosing

Reduced radiation exposure

Enhanced image quality

Improved satisfaction

Faster diagnosis

Improved access to care

Improved reporting

AI in radiology

AI can more reliably detect diseases at early stages, preventing complications and dramatically improving patient outcomes
AI-based radiology tools provide decision support by automatically prioritizing scans based on case severity, saving clinicians’ time and allowing them to provide timely patient care
Most radiology AI tools can detect abnormalities more accurately than human radiologists, increasing patients’ chances for recovery
AI dose optimization systems can help reduce the radiation level patients and radiographers are exposed to during a scan.
AI can help reduce the amount of radiation exposure by providing more accurate images with less need for repeated imaging
AI can help improve the image quality of medical scans, facilitating abnormalities detection and diagnostics
AI-powered radiology tools can help improve patient satisfaction by providing faster and more accurate diagnoses
AI can help speed up the diagnosis process, allowing patients to receive treatment more quickly
By improving patient throughput and making decisions without human involvement, AI can democratize access to radiology worldwide
Most AI-powered radiology tools automatically produce error-free and standardized reports, saving time and streamlining workflows
Early detection

AI can more reliably detect diseases at early stages, preventing complications and dramatically improving patient outcomes

4 challenges with AI in radiology & potential solutions

Aligning medical guidelines with AI outputs

Challenge

Solution

Most current AI applications in radiology provide estimates of how likely a certain patient is to have complications based on radiological imaging. For example, an AI system concludes that a breast lesion of a certain patient has a 10% chance of being malignant. A radiologist could opt for a biopsy, but the AI system may not understand the severity of the problem and deem a 10% chance of cancer insignificant for conducting a biopsy.

Developers and medical professionals must work in close collaboration. Insights from medical professionals can improve the performance of AI-based image processing and decision-support solutions.

Human reluctancy

Challenge

Solution

Designing accurate AI algorithms is as challenging as understanding how to integrate AI systems in daily radiology workflows. The roles and responsibilities of radiologists can change. Still, it is unlikely that algorithms will make fully autonomous clinical decisions, regardless of their implied accuracy and effectiveness.

Healthcare providers should ensure that human-machine collaboration is effective and human professionals keep the lead in the decision-making.

Poor IT infrastructure

Challenge

Solution

Despite many use-cases of AI in radiology, many healthcare providers haven’t even started the digital transformation. Their research and patient data is siloed, security measures are outdated, hardware requires upgrades, and systems lack interoperability. Incorporating AI in such a setting can create more hurdles.

Combating these problems comes down to gradual changes in IT infrastructure, preferably with support from an experienced third-party vendor. Healthcare organizations can start their AI adoption journey with the adoption of image management and PACS systems for improved image quality and easier retrieval.

Data quality

Challenge

Solution

The lack of high-quality labeled datasets is a universal problem across sectors and industries, and radiology is no exception. Gaining access to clear and labeled imaging data for training medical AI is not easy.

Given that many healthcare providers are currently in the midst of digital transformation and the demand for quality datasets is growing every day, it is only a matter of time before the majority of datasets will adhere to high-quality standards.

Want to add AI-based technologies into your radiology practice?

Contact us

AI methods in radiology

AI has become synonymous with machine learning (ML), which encompasses many techniques used to create predictive models, including deep learning models. Today, the most popular way of mimicking human decision-making is artificial neural networks (ANN), which is a certain type of very flexible deep learning model. 

ANN is inspired by the design of biological neural networks. In other words, this model tries to mimic the network of neurons that comprise the human brain so that it can solve problems similar to the way a human would. 

Currently, the convolutional neural network (CNN), which is a class of ANN, is the most popular model for analyzing visual data, making it the most important component of AI in radiology.

Input layer
Hidden layer 1
Hidden layer 2
Output
Scheme title: CNN operation sample Data source: journals.sagepub.com—Artificial intelligence in radiology: friend or foe? Where are we now and where are we heading?

Implementation tips for AI in radiology

Build a multidisciplinary team

While input from field experts is always critical to the development of AI solutions in virtually any industry, medical imaging calls for radiologists to have a much more decisive and significant role in the development of AI-based radiology solutions. Engineers and data scientists lack an understanding of essential radiological and anatomical concepts, which is crucial to creating AI systems applicable in radiology. On top of that, radiology is a very nuanced medical field, where therapeutic methods tend to change every few years. It’s highly advisable to involve radiologists in every phase of development.

Gain support from leadership

Healthcare providers should get key decision-makers and executive leadership on board with the implementation of AI tools. According to the latest report by Deloitte, healthcare providers who succeeded in AI initiatives put leaders at the core of AI transformation. These leaders should clearly communicate with the radiology department on how new tools will impact their daily workflows.

Prioritize AI explainability

In a recent study, radiologists received chest radiographs and diagnostic advice from human experts and from AI systems. The radiologists rated AI advice as lower quality, which highlights the persistent 'AI black box' problem. In healthcare, where decisions can directly influence one's well-being, it is paramount for radiologists to understand the underlying decision-making of AI systems. Most AI radiology tools use deep learning models, and system interpretability and transparency become crucial. While deciphering deep learning models is a universal task across many AI applications, developers are responsible for minimizing bias and variability and making models as explainable as possible. It comes down to continuous algorithm validation, dataset audit, evaluation of its performance metrics, standardized documentation, and routine bias assessment.

Add AI into your radiology practice

Add AI into your radiology practice

AI-based tools are already used in different fields of radiology, and their role will increase moving forward. The validation of existing AI applications and development of new ones for solving common radiological problems requires effective collaboration between radiologists, data scientists, and engineers.

AI will make the next two decades transformational for healthcare and radiology in particular. Our professional AI consultants can help you implement AI-based technology into your radiology practice.

Add AI into your radiology practice

FAQ

Will AI take over radiology?

AI can free radiologists from mundane tasks and improve their efficiency, but human involvement will remain critical to solving complex problems that call for rapid change. Similar to the advent of autopilot in aviation, AI will augment radiologists’ work rather than replace it. Curtis Langlotz, a radiologist at Stanford, said: “AI won’t replace radiologists, but radiologists who use AI will replace radiologists who don’t”.

What is the future of AI in radiology?

With the growing acceptance of AI among medical professionals and regulatory bodies, this technology will take over radiologists’ repetitive tasks and become their essential reference tool.

When was AI introduced in radiology?

The first use of AI in radiology dates back to 1992 and was used to detect microcalcifications in mammography. Arterys, the first AI-based FDA-approved medical imaging solution, was certified in 2017.

AI services and solutions

Service

AI services and solutions

Explore our range of AI services, along with key AI use cases, related technologies, and adoption guidelines.

EMR/EHR software development:
a comprehensive guide

Service

EMR/EHR software development: a comprehensive guide

Discover the major capabilities and use cases of EHR systems and learn how we can support your organization with EMR/EHR software development services.

Healthcare data analytics services: 
key types, features, and applications

Service

Healthcare data analytics services: key types, features, and applications

Discover how our healthcare data analytics services can enhance patient loyalty while growing revenue and learn which solutions fit your business goals.

Custom healthcare software development: solutions & best practices

Service

Custom healthcare software development: solutions & best practices

Entrust your medical software development to a company with 20+ years of experience in healthcare IT. Compliance and security guaranteed.

NLP in healthcare: fostering medical digitalization

Insights

NLP in healthcare: fostering medical digitalization

NLP in healthcare provides a wide array of tools to assist medical professionals and improve patient experience. How? Let's find out.

Digital transformation in healthcare: today’s trends and their positive effects

Insights

Digital transformation in healthcare: today’s trends and their positive effects

We look into digital transformation in healthcare and explore its positive impact on care delivery and waste reduction.