Artificial Intelligence in Pulmonology: threat or opportunity?
Our world has changed drastically in the past decades with the advancement of new technologies. We have seen revolutions that can impact many different areas of our lives: self-driving cars became a reality, computers are beating world champions in chess, navigation apps show us the fastest route… All these technologies have one thing in common: artificial intelligence (AI). What is artificial intelligence? How is it used in medicine? Can pulmonology professionals benefit from the technology?
What is Artificial Intelligence?
AI is the science of making intelligent machines, more specifically intelligent software. The goal is to use this intelligence to augment the abilities of people and enable us to accomplish more and eliminate tedious and repetitive tasks.
How does a computer become intelligent? We as humans often learn from examples, from various experiences, and from repeated practice. Repeated practice and examples are key for AI as well. We expose the system to examples of behavior that we want to achieve. The system will learn from that data what to do or what output to provide in another similar case.
AI in Medicine
There are great examples of where AI is used in medicine already today: mining medical data, discovery of new drugs, robotic surgery, precision medicine… One of the most common applications is diagnosis and screening to detect or classify anomalies. In radiology we see an exponential growth in use of AI to support a more accurate and early diagnosis based on imaging data, but also other medical fields have adopted some very successful AI technologies.
AI in Pulmonology
In the last 10 years the number of scientific publications on Asthma has doubled, the number of scientific publications on AI in respiratory medicine has become 100 times bigger. We see most applications in three areas: thoracic imaging (e.g. automated reporting or triage of chest radiographs), histopathology (e.g. discrimination of lung cancer carcinoma), and applications in physiological measurements.
AI researchers have a habit of creating a competition between humans and computers: a chess grandmaster against an AI-based chess algorithm, many radiologists against an AI-based image screening algorithm… This is excellent to validate if artificial intelligence is well made for a specific task.
However, the real benefit for the patients is in teamwork. As in a great team, the best results are obtained if each team member can focus on its strengths. The same is true with AI and human doctors. Computers are excellent in following rules, doing repetitive tasks, classifying data patterns. That is where they can serve doctors to help them in their daily clinical practice. To allow doctors to focus on where they are indispensable: in clinical decision making that considers the full patient and all the human aspects on top of the data, and in communicating this to the patients at their level so each patient feels confident that they have the best treatment option for them.
AI-assisted Pulmonary Function Test Reporting
Pulmonary function tests (PFTs) are one of the most important tools to evaluate the respiratory system. The interpretation is traditionally an important aspect of the expertise of respiratory specialists. However, many factors contribute to a large variation between experts: the large volume of patients and high workload, the changes in reference equations over the past years, the recent changes in ERS/ATS interpretation guidelines, the differences between different national and international guidelines, the difference in methods and experiences between different institutions. These variations can make it hard in some cases to achieve an accurate diagnosis very early on in the diagnostic process.
Therefore, a group of researchers from the university of Leuven decided to develop a software that would automate PFT reporting.
The software makes sure that all PFT reporting is consistent between colleagues and always in line with the guidelines. All while saving residents hours of typing work and making (preliminary) reports available immediately after testing. In addition, the software leverages the power of AI to support an early and accurate diagnosis.
The most recent ERS/ATS technical standard on interpretive strategies for routine lung function tests state “AI based software may also provide more accurate and standardized interpretations and may serve as a powerful decision support tool to improve clinical practice.”
After a successful implementation, ArtiQ was founded to commercialize the technology and bring it to hospitals across the globe. Since the start, the technology is used over 130 000 times in daily clinical practice.
PFTs provide extensive numerical outputs, and their patterns need to be interpreted within a clinical context to come to a diagnosis. ArtiQ.PFT serves as a second opinion for us. The software is particularly helpful in recognizing rare but specific patterns.
– Prof. Dr. Wim Janssens