PracticeUpdate Conference Series: ERS 2018

AI Improves Testing and Diagnosis of Lung Disease Artificial intelligence (AI) can help in the interpretation of respiratory symptoms. A rtificial intelligence (AI) has proven to be more consistent and accu- rate in interpreting respiratory test results and suggesting diagnoses than lung specialists. Interpretations were measured against gold standard guidelines from the ERS and American Thoracic Society. The expert panel considered patients’ medical histories and results of all pulmonary func- tion tests and any additional tests before agreeing on the correct interpretation and diagnosis of each patient.

Results of this exploration of AI among 120 pulmonologists from 16 hospitals were reported at ERS 2018. Marko Topalovic, PhD, of the Catholic University of Leuven inBelgium, explained that pulmonary function tests provide extensive numerical output patterns that are difficult for the human eye to under- stand, but computers can easily manage these data. Compared with diagnosis by pulmonologists, diagnosis by AI wasmore accurate in twice as many cases, which demonstrates that AI can be used as a second opinion for pulmonologists in the assessment and diagnosis process. He also noted that interpreting pul- monary function tests and diagnosing respiratory diseases is not easy for pulmonologists. To be accurate, more information and additional tests are required. However, AI-based software can be used as a powerful decision support tool for improving clinical practice. Doctors have provided very positive feedback, especially based on the assistance with identifying difficult patterns of rare diseases. According to Dr. Topalovic, the belief is that AI can be used to empower doctors to make interpretations and diagnoses earlier. AI will not replace doctors, how- ever, because doctors can see a broader perspective beyond what is presented by pulmonary function tests. AI can be used to augment the ability to accomplish more and reduce errors and redundant work. He also noted that we currently use computers for flying planes, driving cars, and guarding security. AI can be used to provide high-quality pulmonary function test interpretation regardless of location or medical coverage. The acceptance by the medical community will determine whether AI will be used widely in clinical applications. Dr. Topalovic and coinvestigators used historical data from 1430 patients from 33 Belgianhospitals. Thedatawereassessed by an expert panel of pulmonologists.

Dr. Topalovic explained that quality data is the most important factor to consider when training the AI algorithm. All results of the pulmonary function tests, additional tests, and medical information were reviewed by an expert panel who agreed on the final diagnosis. These diagnoses were used to develop the algorithm for training the AI prior to incorporating it into clinical practice at the University Hospital Leuven for validation. Ensuring that the algorithm could recognize patterns of up to nine different diseases was challenging. Then, 120 pulmonologists from 16 European hospitals in Belgium, France, The Netherlands, Germany, and Luxembourg performed 6000 interpre- tations of pulmonary function test data from 50 randomly selected patients. The AI examined the same data. Results of both were measured against gold standard guidelines the same way as during algorithm development. The researchers found that interpretation of pulmonary function tests by pulmonol- ogists matched guidelines in 74% (56% to 88%) of cases. Software interpretations based on AI matched the guidelines per- fectly (100%).Pulmonologists were able to diagnose the primary disease correctly in 45% (24% to 62%) of cases. AI diagnosed correctly in 82% of cases. Two large Belgian hospitals are using the AI-based software to improve inter- pretations and diagnoses. The AI-based software confers superior performance and may provide a powerful decision support tool to improve clinical practice. The next step will be to get more hospitals to use this technology and investigate transferring the AI technology to primary care, where the data will be captured by general practitioners to help them diagnose and refer correctly.

randomized, placebo-controlled, double- blind trial. Evidence-based 2011 clinical prac- tice guidelines for the diagnosis and management of idiopathic pulmonary fibrosis defined the disorder as pro- gressive fibrotic lung disease limited to the lungs. Interstitial pulmonary fibrosis occurs in adults without attributable sys- temic disease and environmental factors. The guidelines include precise imaging and histopathological criteria for pat- terns of usual interstitial pneumonia (the hallmark feature of interstitial pulmonary fibrosis in the lung). Since then, new evidence has changed the landscape of treatment for interstitial pulmonary fibrosis. Idiopathic pulmonary fibrosis is a fibro- proliferative interstitial lung disease of unknown etiology that results in a pro- gressive loss of lung function and median survival of 3–5 years. Dr. Wung concluded that circulating ANA is prevalent among patients with idio- pathic pulmonary fibrosis. Its overall clinical significance remains to be determined, but patients with idi- opathic pulmonary fibrosis with high-titer ANA likely represent a unique subset of the overall population with idiopathic pulmonary fibrosis.

www.practiceupdate.com/c/73941

www.practiceupdate.com/c/73944

3

ERS 2018 • PRACTICEUPDATE CONFERENCE SERIES