The focus on preventative healthcare has grown exponentially, with early diagnosis taking top priority. Practitioners are facing mounting pressure to deliver accurate and timely diagnoses. None more so than radiologists, who handle an ever-increasing volume of medical images in a day. While the number of hours remains the same. Dr Henno Schoombee, a radiologist at SCP Radiology, answers questions about AI in radiology, the future of digital imaging and the impact on radiologists and patients.

Is AI in radiology a new development?

We have been using AI a lot longer than most people realise and it’s advancing at a rapid rate. I believe that after 10 years we are at the bottom of a growth curve which is about to take off. Earlier methods produced subhuman performance, while recent deep learning algorithms match or are superior in certain areas.

From the early days of X-ray imaging in the 1890s to more technologically advanced PET, CT, and MRI images – these diagnostic and investigative tools remain the pillar of medical treatment. The inclusion of AI tools will take imaging to another level.

What is AI in radiology?

It is best defined as a collection of algorithms, machine learning tools, sophisticated neural networks and computer aided detection (CAD) systems that are changing how radiology services are delivered. AI is transforming image acquisition, post-processing, workflow optimisation, image interpretation, report creation and results communication.

In addition, AI-powered Natural Processing Language (NPL) gives computers the ability to understand text and spoken words in much the same way as human beings. NPL tools can extract relevant information from radiology reports, enhance documentation accuracy and facilitate data mining, for research purposes.

What is deep learning in AI?

AI models can analyse large volumes of medical images with unprecedented speed and accuracy. These machine learning algorithms excel at recognising patterns and abnormalities, allowing radiologists to detect diseases such as cancer, strokes and fractures with greater precision. Deep learning can interpret thousands of existing images, together with their diagnoses, to apply in future diagnoses. It is by no means perfect but has a significant impact.

On a personal level, the most important reward is improving care – it helps us find new cancers and patients with good results after chemotherapy, and detect subtle but dramatic injuries. It also improves patient care, indirectly, by reducing the workload for radiologists. AI will go a long way towards taking away mundane tasks, so we can focus on the more specialised tasks.

How is AI beneficial to both patients and radiologists?

• AI offers significant advancements in workflow optimisation and efficiency.
• Automated image triage systems can prioritise urgent cases, ensure prompt attention and expedite patient care.
• AI algorithms assist in routine tasks such as image pre-processing, segmentation, and annotation, saving valuable time for radiologists.
• In medical imaging it assists radiology in Picture Archiving and Communication Systems (PACS). This improves structured reporting, auto detect injuries and diseases and will automatically pull across relevant prior exams and patient data.

.

.

Are the results more accurate?

One of the most important applications is faster, more accurate results. Ultimately, we are aiming for sufficient time for radiologists to spend interpreting data, rather than having to collect this manually, which is extremely time consuming. For example, if AI can compare a nodule or tumour three or six months from the initial imaging, it can link the previous investigations automatically and make comparisons so that the radiologist can see the changes and interpret them.

That is how clinician-AI collaboration is crucial.

Where in AI is a tangible difference?

Interpretive imaging helps with diagnosis by using algorithms running in the background in a particular area. These algorithms can detect cancers earlier and radiologists can then decide if it’s worth investigating such as:

• Mammograms

We currently use it for mammography to detect lumps before they may be visible. This allows the radiologist to decide if a biopsy is necessary or not

• Diagnosing collapsed lungs

It is useful in determining even a marginally collapsed lung (Pneumothorax) which is particularly difficult to detect, as well as embolisms in the pulmonary arteries, tumours in the chest and heart as well as brain haemorrhages

• Fractures

A busy radiologist may miss a hairline fracture but AI will be able to identify an anomaly.

The important task of measurements

One of the greatest time savers is automatic measuring by AI.  Radiologists are still required to measure a number of angles on images. I believe this time can be used more productively but, it’s important to note that even though these measurements are incorporated into the report – findings must be validated by a radiologist. 

The burning question… Will AI ever replace radiologists completely?

Never say never. But I don’t believe so. More accurately I would say AI will replace the mundane tasks so that our skills are used more in interpretative radiology.

We see AI as a tool to improve patient care, not take it over.

What is the future?

In short, AI has the potential to revolutionise the field of radiology, empowering radiologists with powerful tools to improve diagnostics, streamline workflows and enhance patient care.

Dr Henno Schoombee is a radiologist at SCP Radiology, an independent radiology practice.  SCP Radiology has been providing medical imaging services in the Western Cape since 1950.

LISTEN: Dr Jean de Villiers talks about AI in radiology on CapeTalk and RSG.