Predicting Recurrent Pneumonias and Exacerbations in Bronchiectasis

Project descpription:

Endometriosis affects approximately 10% of fertile women, causing significant pain and infertility, yet non-invasive diagnostic tools remain inadequate. […] This project aims to address these challenges by developing a novel diagnostic approach using Positron Emission Tomography (PET). PET offers higher resolution and sensitivity for detecting and characterizing tissues. Recent research has shown that endometriotic lesions are rich in relaxed fibronectin (Fn), particularly in the stroma and areas of reactive fibrosis. Preliminary studies have successfully stained Fn in endometriotic tissues, suggesting it could serve as a promising target for PET imaging. The project proposes a detailed analysis of several extracellular matrix proteins, including relaxed fibronectin, fibroblast activation protein (FAP), matrix metalloproteinases, and integrins, using immunohistochemistry. These markers will be correlated with MRI and ultrasound findings, as well as intraoperative observations and histopathology results. The goal is to identify an optimal target for PET imaging that can differentiate active, painful endometriotic tissue from scar tissue, thereby improving preoperative lesion localization and surgical planning. This approach could potentially allow for earlier, more accurate diagnosis, better patient management, and improved pain relief after surgery.

Preliminary results:

The AERO project develops clinical and AI-based tools to identify bronchiectasis patients at increased risk of recurrent infectious exacerbations or pneumonias. During the first 12 months, a CT data foundation with more than 3,000 available scans was established, including 424 fully annotated and quality-checked scans from 410 patients. Clinical CT phenotyping showed mucus plugging as the strongest imaging marker of high-risk bronchiectasis (65.4% vs. 15.8%; OR 10.06, 95% CI 5.14–19.67; p<0.01). Consolidations (53.8% vs. 20.5%; OR 4.52, 95% CI 2.41–8.50; p=0.01) and ground-glass opacities (46.2% vs. 23.5%; OR 2.79, 95% CI 1.50–5.20; p=0.01) were also significantly associated with high-risk status. In parallel, a reproducible 3D CT pipeline using lung masking and a Vista3D foundation-model encoder achieved promising performance for predicting recurrent infectious exacerbations in patients with bronchiectasis, with balanced accuracies of 0.886 in cross-validation, 0.882 in independent out-of-distribution testing, and 0.834 in longitudinal patient-level analysis. Together, these findings indicate that clinically visible disease patterns on routine chest CT can be systematically captured and combined with AI-based analysis to support bronchiectasis risk stratification. The next phase will focus on external validation, dataset expansion, longitudinal modelling, clinical integration and threshold calibration.

About Tician Schnitzler:

Dr. med. Tician Schnitzler is a radiology resident and postdoctoral researcher at Kantonsspital Aarau, specializing in thoracic imaging and artificial intelligence (AI) applications in medical imaging. His expertise has been shaped by a research fellowship at the University of California, San Francisco (UCSF), where he focused on integrating AI into clinical radiology. He holds an MD from RWTH Aachen University, Germany, and is currently pursuing a Master’s in Biomedical Informatics and Data Science at the University of Mannheim to further enhance his skills in data-driven medical research.


Dr. Schnitzler will lead the project "Predicting Recurrent Pneumonias and Exacerbations in Bronchiectasis: Clinical and Imaging Phenotypes for Risk Stratification and Algorithm-Assisted Management." In this role, he applies his specialized skills to develop predictive models that integrate clinical data with chest CT imaging, aiming to identify bronchiectasis patients at high risk for pneumonia and exacerbations. This project represents a unique collaboration between Kantonsspital Aarau and the Paul Scherrer Institute (PSI) with the goal to advance AI-driven tools that enhance patient management—particularly in the context of Switzerland’s upcoming national lung cancer screening program.