Deep learning in pulmonary drug delivery

Abstract

Pulmonary drug delivery is essential for lung disease therapy like inhalable SARS-CoV-2 vaccines are developing to combat the current COVID-19 pandemic. To prompt novel aerosol delivery strategies, high-resolution visualization, reconstruction, and quantification of inhaled drugs within the complex 3D architecture of the lung is crucial, yet, elusive even in preclinical models. This study combines cut-edging imaging technology with artificial intelligence (AI) approaches to accomplish this in a mouse model. Tissue-cleared light sheet fluorescent microscopy of murine lungs enables co-mapping the lung geometry and drug biodistribution at cellular resolution following pulmonary drug delivery via different routes such as intranasal and oropharyngeal aspiration, intubated instillation as well as ventilator-assisted and nose only aerosol inhalation. We further employ AI and deep learning pipelines (convolutional neural network, CNN) for the segmentation and reconstruction of the bronchial airway tree from the trachea, via all generations of bronchi and bronchioles to the terminal bronchioles. This clearly shows the more central and patchy deposition pattern of the three bulk liquid application methods as compared to the two aerosol inhalation methods. Most notably, inhalation of micron-sized droplets results in an extreme hotspot aerosol deposition pattern in the proximal acinar region (>10-fold dose/area increase), which has been predicted by numerical computational lung deposition models, but not been visualized yet. Moreover, we provided region-resolved dose distribution and dosimetry maps in each individual order of airways in an automatic way, shedding light on targeted drug delivery and pulmonary precision diagnostics and medicine.