Artificial intelligence analysis of chest CT scans has generated six different emphysema subtypes that may help drive more precise diagnosis and personalised treatment of pre-COPD and COPD, US researchers report.

Study investigators trained a custom-designed machine learning algorithm to process the texture and anatomical location of emphysematous regions on 2,853 chest CT scans of participants in the Subpopulations and Intermediate Outcome Measures in COPD Study (SPIROMICS).

Ultimately, this identified six different emphysema subtypes, each with distinct symptoms, physiology and prognosis, among which three also had linked genetic associations.

According to the authors, the findings suggested potential “paths to specific diagnosis and personalised therapies in COPD and pre-COPD”.

The two most common subtypes were found to predict incident airflow limitation in people who did not have COPD, thus “improving the specificity of pre-COPD,” they noted in their paper published in Thorax.

The most common subtype in SPIROMICS was the combined bronchitis-apical subtype, which was also linked with chronic bronchitis, accelerated lung function decline, hospitalisations, deaths, and a gene variant near DRD1 implicated in mucin hypersecretion, followed by the diffuse subtype, which was also associated with respiratory hospitalisations and deaths, and lower body weight.

The third most common subtype was linked with age only; the fourth and fifth “visually resembled combined pulmonary fibrosis emphysema and had distinct symptoms, physiology, prognosis and genetic associations,” and the sixth visually resembled vanishing lung syndrome, according to the paper.

“All resembled early COPD subtypes, which are ignored in contemporary guidelines, and provide precise CT-defined subtypes, some of which suggest avenues to personalised medicine”, the authors said.

However, they stressed that additional studies are necessary to determine whether the implicated genes are causal and whether drugs targeting identified pathways will successfully generate personalised treatment approaches for pre-COPD and COPD.