Bronchiectasis disease isn’t a single disease – it’s a final common pathway for many. From primary ciliary dyskinesia and post-infectious damage to chronic inflammation and mucus clearance dysfunction, the underlying causes differ widely, and so do the structural manifestations on CT.^[1]

This heterogeneity within and between patients makes it exceptionally hard to run clinical trials. Two patients may both carry the same diagnosis yet have fundamentally different structural disease patterns: one with diffuse airway wall thickening, another with severe dilation and mucus impaction. Traditional imaging reads simply don’t capture this variation with the accuracy needed to stratify patients or measure treatment response.

Worse yet, our go-to endpoints like exacerbation rates or FEV₁ (Forced Expiratory Volume in one second – a crucial measurement in pulmonary function testing) often miss the mark. FEV₁ may remain unchanged while underlying structural abnormalities are progressing. Exacerbations are erratic, difficult to define, and are influenced by external triggers and reporting bias. And visual CT scoring, even when performed by experienced clinicians, remains subjective and time-consuming, with high variability across readers and sites. In essence, we were trying to navigate a bronchiectasis disease… without a proper map.

And here we are today, with AI bringing a whole new dimension to image interpretation, not only advancing but holding the potential to transform treatment of lung diseases. The emerging solutions come not from a new drug or device, but from a new way of seeing.

Using deep learning algorithms, at Thirona we trained our LungQ® software to extract precise, quantitative measurements from CT scans – automatically and reproducibly. We weren’t just looking at the airways anymore, but measuring them down to the submillimeter, and we can now quantify:

• The degree of bronchial dilation, generation by generation.
• The extent of wall thickening across lobes and segments.
• The volume, number, and location of mucus plugs.
• The pulmonary blood volume distribution and its response to treatment.

Analyzing patterns across various bronchiectasis patient populations in multiple studies, we find patients with extensive wall thickening and minimal dilation^[2], an image that would often go unnoticed in traditional visual reads. Others show diffuse dilation with little thickening. Mucus plug burden varied dramatically, from none to hundreds per lung. These structural phenotypes are no longer anecdotal observations – they become measurable, classifiable, and clinically actionable.

In a study of over 600 bronchiectasis patients from the EMBARC registry, we can see it clearly: this disease does not follow one path, and it certainly doesn’t respond to treatment in a uniform way.^[3]

What strikes me most is how these AI-derived imaging metrics are redefining the way we can conduct clinical trials in bronchiectasis disease, and across respiratory disease more broadly.

By moving from subjective visual reads to automated, quantitative metrics, we gain a level of reproducibility and sensitivity that traditional endpoints simply can’t offer. For instance, structural metrics such as bronchial wall thickness or mucus plug number can reveal meaningful treatment effects even when spirometry remains unchanged^[4], which brings a critical advantage in phase 2 and 3 studies where sensitive and accurate signals matter.

Moreover, this level of granularity allows us to track regional disease activity, at lobar or even segmental level, and understand where therapies are acting, and where they are not.^[5] These insights turn imaging into a source of functional biomarkers, supporting both patient stratification and therapeutic mechanism-of-action studies.