Subscribe
Share
blog post background
Strategy

Artificial intelligence in radiology - use cases and trends

By Nadejda Alkhaldi, Innovation Analyst
Published on

The radiology department of an average healthcare facility is likely searching for improvements. Even before the pandemic, 45% of radiologists experienced burnout at some point in their careers. They felt overwhelmed with the administrative burden and the large number of images they had to check manually, which could reach up to a hundred scans per day. Additionally, radiology practice is lacking non-invasive methods for tissue classification. Invasive procedures take time and cause stress for patients.
Luckily, AI healthcare solutions are coming to the rescue. Just take a look at these numbers:

  • By implementing AI for image reconstruction, GE Healthcare could scan 35 patients per day instead of 25
  • By using AI for mammogram reading, researchers in Sweden reduced radiologists’s reading load by 44%

So, what can we expect from artificial intelligence in radiology, and how do healthcare leaders already use this technology at their facilities?

What is AI in radiology?

Artificial intelligence algorithms excel at analyzing enormous amounts of visual data quickly and executing complex tasks that typically require human intelligence.

When it comes to radiology, different subtypes of AI can assist doctors on different levels. For instance, deep learning and computer vision can analyze medical images and come up with preliminary diagnosis. They have an outstanding “eye for detail” and can spot slight line variations that radiologists wouldn’t notice. A more recent AI subtype, generative AI, can learn to produce original content that will easily pass for human creation. These models can synthesize medical images and help generate drafts of radiology reports.

As more clinics with radiology departments are looking to adopt this technology, its value increases. The global AI in the medical imaging market was valued at $0.98 billion in 2023, and it’s estimated to exceed $14 billion by 2032.

AI-in-Radiology
Source

Let’s see what benefits this technology can bring.

Benefits of AI in radiology

There are four key benefits to deploying AI in radiology:

  1. Detecting diseases in early stages. Artificial intelligence models view scans differently and can catch early disease onset signs that a human eye can’t see. For example, a research team in Boston built an AI tool that can predict if a person will develop lung cancer in the coming year with an accuracy of 86% to 94%. Using such tools could revolutionize cancer treatment.

  2. Improving patient outcomes. In addition to predicting diseases, AI can support radiologists in screening medical scans and detecting tumors. Taking breast cancer as an example, a recent study shows that AI-assisted mammogram examination improves cancer detection rate by 20%.

    Also, embedding AI into scanners accelerates the scanning process, reducing patient exposure to radiation. For example, GE Healthcare reported a 30% reduction in scanning time with such machinery.

  3. Reducing the amount of stress that radiologists experience on the job. Being responsible for detecting cancer and other life-threatening diseases is already stressful enough. But radiologists are also overwhelmed with administrative duties, such as producing reports. AI in radiology can offer a second opinion on diagnosis and help with administrative tasks. For example, a tech company HOPPR developed a foundation model called Grace that can assist radiologists in reading medical scans, transcribe their comments, and potentially fill in radiology reports with this information.

  4. Supporting education, research, and training. Generative AI in radiology can synthesize realistic medical scans to enhance existing datasets and facilitate research. It can also simulate different disease development scenarios for training purposes.

Top applications of artificial intelligence in radiology

Computer-aided detection (CAD) was the first application of AI in radiology. Traditional CAD systems use predefined algorithms for recognizing patterns and can only identify abnormalities that they were explicitly trained to detect. These systems do not have the capability for autonomous learning, meaning any new functionality must be manually programmed.

Since the advent of CAD, AI has evolved significantly and now offers a broader range of support for radiology departments. Some of the modern medical digital image platforms enable users to manage different types of images, manipulate them, connect to third-party health systems, and more.

So, how can AI-powered medical imaging solutions help radiologists?

1. Classifying brain tumors

Brain cancer, along with other types of nervous system cancers, is the 10th leading cause of death in the US.

Conventionally, brain tumor patients and their surgeons were oblivious to how the surgery would go, as the doctor wouldn’t know which kind of tumor was there and what treatment the patient would have to undergo. And this element of surprise could be fatal if the surgeon didn’t successfully adjust the treatment process.

Introducing AI in radiology to this mix can either give the surgeon a heads up before the operation or accelerate tumor classification during surgery.

Real-life examples:

Researchers at UMC Utrecht, the Netherlands, built an AI-powered tool that can classify brain tumors during surgery. This software can study the tumor’s DNA and determine its type and even its subtype, giving the surgeon instructions on how to proceed. If the tumor is aggressive, doctors will remove more brain cells when in doubt. Otherwise, they can operate sparingly without risking leaving diseased cells inside.
This solution was tested on frozen brain samples and correctly diagnosed 45 of them. For the other five, the model refused to give a recommendation.

Another fascinating example comes from the Shenzhen Institute of Advanced Technology. The Chinese research team developed a deep learning model that can classify diffuse gliomas by analyzing whole-side images (WSIs) without the need for molecular testing.

2. Detecting hidden fractures

The FDA started clearing AI algorithms for clinical decision support in 2018. Imagen’s OsteoDetect software was among the first the agency approved. This program uses AI to detect distal radius fractures in wrist scans. The FDA granted its clearance after Imagen submitted a study of its software performance on 1,000 wrist images. The confidence in OsteoDetect increased after 24 healthcare providers using the tool confirmed that it helped them detect fractures.

Another task where AI in radiology shines is spotting hip fractures. Traditionally, radiologists use X-rays to detect this type of injury. However, such fractures are hard to spot as they can hide under soft tissues.

Real-life example:

A research team tested how YOLOv4-tiny AI algorithm performs on hip fracture detection. They trained the model on 900 medical images and tested it on 100 scans. The model performed with an accuracy of 95%, surpassing first-year medical residents and matching the precision of experienced radiologists.

3. Recognizing breast cancer

Breast cancer is the second leading cause of death among women in the US.

Despite the severity of this disease, doctors miss up to 40% of breast lesions during routine screenings. At the same time, only around 10% of women with suspicious mammograms appear to have cancer. The other 90% have to struggle with frustration and depression and even undergo invasive procedures while doctors further investigate their condition.
Here’s how AI in radiology can change that.

Real-life example:

In a recent study conducted in Sweden, researchers wanted to compare the performance of an AI-powered breast cancer detection solution with that of human radiologists. This team used medical scans of a group of women with an average age of 54. Each scan was examined by two radiologists and by the AI tool. As a result, AI supported screening detected anomalies in 244 patients, while radiologists recalled 203 women for further investigation. Out of the 41 patients spotted by AI and missed by the radiologists, 19 actually proved to have invasive cancer.

The researchers also reported that false-positive rate was 1.5% for both AI and radiologists, which contradicts the common belief that using AI in radiology results in a high false-positive rate.

4. Detecting neurological abnormalities

Artificial intelligence in radiology has the potential to diagnose neurodegenerative disorders such as Alzheimer’s, Parkinson’s, and amyotrophic lateral sclerosis (ALS) by tracking retinal movements. This analysis takes around 10 seconds. Another approach to spotting neurological abnormalities is through speech analysis, since Alzheimer’s changes patients’ language patterns. For instance, people with this disorder tend to replace nouns with pronouns.

AI helps doctors identify which patients with mild cognitive impairment will go on to develop degenerative diseases and how severely their cognitive and motor skills will decline over time. With this knowledge, the endangered patients have the chance to arrange for care facilities while they still can.

Real-life example:

One notable AI-based radiology solution for detecting neurological abnormalities is the model developed by researchers at Georgia State University’s TReNDS Center. This AI model is designed to analyze functional Magnetic Resonance Imaging (fMRI) data to detect various neurological disorders, including Alzheimer’s disease, schizophrenia, and autism spectrum disorder. The AI was trained on fMRI data from 10,000 individuals without known neurological disorders and then tested on data from 1,200 individuals with these conditions. By identifying specific patterns and changes in brain activity, the AI model can potentially detect early signs of these disorders, facilitating earlier and more accurate diagnosis.

5. Applications of generative AI in radiology

Gen AI can learn to produce content that looks like it was crafted by a person. This is a relatively new subtype of AI that appeared on the horizon in 2014, after the creation of generative adversarial networks (GANs). Check out our blog to better understand the difference between generative AI and artificial intelligence.

Here is how Gen AI can support the radiology department.

5.1. Improving medical image quality

Generative AI in radiology can significantly improve the quality of medical images by removing distortions and noise and increasing the resolution. It can even reconstruct blurred image regions. For instance, super-resolution GANs can improve the quality of low-dose CT scans by four times while maintaining all the details.

Gen AI can also segment medical images, highlighting organs and abnormalities.

5.2. Generating radiology reports

Deploying generative AI in radiology can help doctors fill in reports and even fully generate the first draft. For example, a team of researchers developed a Gen AI-powered tool that can interpret chest X-rays and deliver radiology reports describing the scan. The team trained this model on 900,000 medical images and reports and asked radiologists to rate the model’s performance. The doctors were pleased with the results and didn’t notice any inconsistencies, including in life-threatening cases, such as pneumonia.

5.3. Synthesizing new medical imaging datasets

Lack of medical data available for AI model training is a well-known challenge of AI implementation in healthcare. Gen AI can help overcome this issue by generating synthetic images with desired properties that closely resemble medical scans.

For example, researchers at Cornell University developed RadImageGAN, a Gen AI-based tool that can produce datasets of high-quality medical scans across 12 anatomical regions and 130 pathological classes. This solution can even generate annotated images.

Challenges you might face when deploying AI in radiology

After viewing the exciting AI applications in radiology, one could assume that the sky’s the limit for this technology. It is indeed very promising, but there are associated challenges that you need to prepare for.

Challenge 1: Development hurdles

There are several concerns associated with AI algorithm development, such as:

  • Availability of training datasets. To function properly, AI in radiology needs to train on large amounts of medical images, but such data is scarce. For the sake of comparison, a typical non-medical imaging dataset can contain up to 100,000,000 images, while medical imaging sets rarely exceed about 1,000 images. With Gen AI, this issue is becoming less critical as the technology can synthesize medical imaging datasets.

  • Data bias. If an algorithm is trained on medical images of a specific racial profile, it might not perform as well on others. For example, software accustomed to white people’s scans will be less precise for people of color. Also, algorithms trained and used at one institution need to be treated with caution when transferred to another organization. A study by Harvard discovered that algorithms trained on CT scans can even become biased toward CT machine manufacturers.

  • Algorithm selection. When it’s time to choose the model, you have three options. You can either purchase a ready-made tool and use it as it is, customize and/or retrain an off-the-shelf solution, or build a model from scratch. When speaking of Gen AI, building a model from the ground up is too expensive and just not feasible. Instead, a generative AI development company can help you select a readily available foundation model and retrain it on your custom data.

Challenge 2: User experience can use a makeover

One of the most common remarks radiologists make is about their rough experience with medical imaging software. It takes many clicks, long waiting times, and a thorough manual study to accomplish even a simple task. Medical programs focus on the technical aspect of performing the job, but their interface is counter-intuitive and not user-friendly.

Challenge 3: Compelling business use cases are hard to identify

One of the hardest barriers to overcome when deploying artificial intelligence in radiology is convincing decision makers that AI is a worthy cause. This technology takes hefty investments upfront, but it will not pay back quickly. Radiologists will need time to learn how to use it.

Suppose you concentrate solely on the radiology department. In that case, AI tools will help radiologists read scans and produce reports faster, which is only a matter of efficiency and does not qualify for such a big investment. The solution here is to broaden your perspective. Almost all departments use medical imaging, so the return on investment won’t be limited to radiology.

Challenge 4: The black-box nature of AI

Algorithms that are built as a black box can’t explain how they come up with their recommendations. And this property is hard to accept when human lives are on the line. So, if it’s impossible to use explainable AI, make sure that human radiologists are responsible for the final decisions and that AI tools merely act as consultants.

Also, many AI models continue learning as they work, acquiring more false logic. For example, algorithms might learn that implanted medical devices and scars are signs of health issues. This is a correct assumption, but then the algorithm might assume that patients lacking these marks are healthy, which is not always true.

Challenge 5: Loose ethics and regulations

There are several ethical and regulatory issues associated with the use of AI in radiology.

  • Unexpected behavior. You can’t predict how the algorithm will act. This is especially true for Gen AI models. For instance, soon after ChatGPT was first introduced, the electronics giant Samsung reported proprietary data leakage as three of its engineers shared some confidential lines of code that the model used for training and mentioned to other users. Now we know better than to share sensitive information with this model, but it wasn’t that obvious in the beginning.

  • Responsibility. Another issue up for debate is who holds the final responsibility if AI led to a wrong diagnosis and the prescribed treatment caused harm. What if the AI model gave a wrong recommendation because it was trained on biased data? Are the AI developers responsible for this?

  • Permissions and credit sharing. The third hurdle is the use of patient data for AI training. There is a need to obtain and re-obtain patient consent and offer reliable and compliant data storage options. Also, if you trained AI algorithms on patient data and then sold it and made a profit, are the patients entitled to a part of it?

You can find more information on how to implement AI in your organization and overcome the challenges above on our blog. We also prepared a free eBook for you on the topic.

The future of artificial intelligence in radiology

Many are wondering how AI will affect radiology and whether it will take over this field and replace human physicians completely. The answer to that is NO. In its current capacity, AI is not powerful enough to solve all the complex clinical problems radiologists are dealing with daily. As Curtis Langlotz, a professor of radiology at Stanford, said,

“AI won’t replace radiologists, but radiologists who use AI will replace radiologists who don’t.”

Healthcare professionals expect AI in radiology to maintain a supporting role, helping doctors with administrative tasks and offering recommendations on diagnosis.

Another interesting usage of artificial intelligence in radiology is augmenting the abilities of doctors in developing countries. For example, researchers at Stanford University are building a tool that will enable physicians to take pictures of an X-ray film using their smartphones. The algorithms underlying this tool will scan the film for tuberculosis and other problems. This app’s benefit is that it works with X-rays and doesn’t require advanced digital scans that are lacking in poor countries. Not to mention that hospitals in these countries might not have radiologists at all.

If you haven’t done that yet, it’s time to enrich your operations with artificial intelligence to take the pressure off your radiologists and improve diagnosing accuracy. Opt for ITRex as your healthcare software vendor. We have extensive experience in AI, and our contribution won’t be limited by building and deploying an AI model. We will also train your staff on proper, compliant usage of artificial intelligence and will help you manage the vast amounts of data that your model will consume and generate.

TABLE OF CONTENTS
What is AI in radiology?Benefits of AI in radiologyTop applications of artificial intelligence in radiology1. Classifying brain tumors2. Detecting hidden fractures3. Recognizing breast cancer4. Detecting neurological abnormalities5. Applications of generative AI in radiologyChallenges you might face when deploying AI in radiologyChallenge 1: Development hurdlesChallenge 2: User experience can use a makeoverChallenge 3: Compelling business use cases are hard to identifyChallenge 4: The black-box nature of AIChallenge 5: Loose ethics and regulationsThe future of artificial intelligence in radiology
Contact ITRex
Contact us
background banner

If you want to learn more about AI applications in radiology and how to overcome deployment challenges, feel free to contact our AI experts.