Artificial intelligence in COPD: Possible applications and future prospects

The cornerstone of artificial intelligence (AI) is its inherent ability to learn from past experience, thus mimicking the human mind and deviating traditional approaches with predefined rules. (machine) learning process involves examples shown algorithm for training purposes performance then assessed new unseen examples. Specifically, during phase, dependencies are defined these used predict outcomes data. Data mining encompassing lower level programming steps in an attempt discover interesting useful patterns large volumes complex Figure 1 depicts schematically relationship among AI, machine data within overall task exploration exploitation. A key distinction between that humans able extrapolate knowledge, therefore, conjecturing associations rules relatively small incomplete amounts data, whereas algorithms, by definition, achieve better as amount increases. Moreover, algorithms assimilate numbers patients, far greater than ones a single physician can manage comprehend entire career. Even though AI branches have been around several decades, it not until recently they grown extraordinary pace healthcare domain. reason this essentially twofold: (i) accumulation increasing complexity via electronic medical records (ii) fact gradually become faster, more effective and, certain extent, becoming explainable. first specialties benefited adoption techniques were imaging oncology, followed neurology cardiology. number publications related respiratory medicine utilize exponentially last few years1 chronic obstructive pulmonary disease (COPD) accounting studies. COPD constitutes ideal target reasons: progressive obscure gene–environment interactions; manifestations fluctuate greatly time, scale dimension; affected patients undergo numerous tests producing clinical, imaging, genomic, metabolomic proteomic spirometries (especially computed tomography [CT] scans) being most integral. It should also be highlighted condition, majority time-course, which adds extra layer has proven counter effectively, identifying trajectories. In literature, applications research span across all aspects disease, is, diagnosing subsequently classifying them into meaningful categories orchestrating their treatment management, eventually capturing progression prognosis disease. terms diagnosis, where spirometric evaluation essential, was proposed2 aiming recognize pattern function (PFTs) yielding perfect accuracy, compared 75% achieved pulmonologists, assign potential diagnosis patient encounter given PFT clinical information, accuracy software pulmonologists 82% 45%, respectively. trained sizable diverse set approximately 1500 validated prospectively collected cohort 50 well-annotated patients. This study highlights two significant factors process: quantity quality crucial. As data-driven, improves sequentially available, assimilating collective knowledge sets. highly dependent on input respective annotation, conforming cliche ‘garbage in, garbage out’. Another critical issue paradox underdiagnosis overdiagnosis disease.3 recent study, deep applied low-dose CTs automated detection COPD, resulting area under curve 0.89, calculated independent test 2000 CT scans.4 Validation external sets contributes unbiased assessment algorithms' provides reliable estimate actual setting. analysis specifically scans, perhaps field expected shine research, volume To end, contribution models invaluable, capture big (in samples, parameters or both) algorithms. subtypes characterization, COPDGene date comprehensive extensive database, 10-year longitudinal genomic diagnosed variable severity5; primary purpose relate phenotypes underlying molecular genetic patterns. importance acquiring analysing time-course order nature Regarding mortality prediction proposed, utilizing 3900 ECLIPSE cohorts.6 Top predictors 6-min walk distance, forced expiratory s % predicted age artery-to-aorta ratio. model outperformed previously existing composite indexes (BODE [Body mass index, airflow Obstruction, Dyspnea, Exercise capacity], BODE modifications ADO [Age, Dyspnoea, Obstruction]) prediction, using fewer predictors, suggesting implication future practice. Overall, lately ‘flooded’ techniques, techniques. An important advantage works same manner every time invoked biased urgent pressure setting often imposed doctors. provide answer specific consideration limited frame, fraction second. Nevertheless, similar imminent future, final decision responsibility remain hands treating physicians, supported computing power algorithm, ultimately targeting benefit patient. Dr Konstantinos Kostikas reports grants, personal fees non-financial support AstraZeneca, Boehringer Ingelheim, Chiesi, ELPEN, GSK, Menarini Novartis; grants NuvoAir; Sanofi, outside submitted work. employee shareholder Novartis Pharma AG 31 October 2018.