Search for: All records

Recurrent respiratory papillomatosis (RRP) is a chronic condition primarily affecting children, known as juvenile onset RRP (JORRP), caused by a viral infection. Antiviral medications have been used to reduce the need for frequent surgeries, slow the growth of papillomata, and prevent disease spread. Effective treatment of JORRP necessitates targeted drug delivery (TDD) to ensure that inhaled aerosolized drugs reach specific sites, such as the larynx and glottis, without harming healthy tissues. Using computational fluid particle dynamics (CFPD) and machine learning (ML), this study (1) investigated how drug properties and individual factors influence TDD efficiency for JORRP treatment and (2) developed personalized inhalation therapy using an ML-empowered smart inhaler control algorithm for precise medication release. This algorithm optimizes the inhaler nozzle position and diameter based on drug and patient-specific data, enhancing drug delivery to the larynx and glottis. CFPD simulations show that particle size significantly affects deposition fractions in the upper airway, emphasizing the importance of particle size selection. Additionally, optimal nozzle diameter and delivery efficiency depend on particle size, inhalation flow rate, and release time. The ML-based TDD strategy, employing a classification and regression tree model, outperforms conventional inhalation therapy by achieving a higher delivery efficiency to the larynx and glottis. This innovative concept of an ML-empowered smart inhaler represents a promising step toward personalized and precise pulmonary healthcare through inhalation therapy. It demonstrates the potential of AI-driven smart inhalers for improving the treatment outcomes of lung diseases that require TDD at designated lung sites. 

Abstract It is challenging to locate small-airway obstructions induced by chronic obstructive pulmonary disease (COPD) directly from visualization using available medical imaging techniques. Accordingly, this study proposes an innovative and noninvasive diagnostic method to detect obstruction locations using computational fluid dynamics (CFD) and convolutional neural network (CNN). Specifically, expiratory airflow velocity contours were obtained from CFD simulations in a subject-specific 3D tracheobronchial tree. One case representing normal airways and 990 cases associated with different obstruction sites were investigated using CFD. The expiratory airflow velocity contours at a selected cross section in the trachea were labeled and stored as the database for training and testing two CNN models, i.e., ResNet50 and YOLOv4. Gradient-weighted class activation mapping (Grad-CAM) and the Pearson correlation coefficient were employed and calculated to classify small-airway obstruction locations and pulmonary airflow pattern shifts and highlight the highly correlated regions in the contours for locating the obstruction sites. Results indicate that the airflow velocity pattern shifts are difficult to directly visualize based on the comparisons of CFD velocity contours. CNN results show strong relevance exists between the locations of the obstruction and the expiratory airflow velocity contours. The two CNN-based models are both capable of classifying the left lung, right lung, and both lungs obstructions well using the CFD simulated airflow contour images with total accuracy higher than 95.07%. The two automatic classification algorithms are highly transformative to clinical practice for early diagnosis of obstruction locations in the lung using the expiratory airflow velocity distributions, which could be imaged using hyperpolarized magnetic resonance imaging. 

Identifying dust aerosols from passive satellite images is of great interest for many applications. In this study, we developed five different machine-learning (ML) based algorithms, including Logistic Regression, K Nearest Neighbor, Random Forest (RF), Feed Forward Neural Network (FFNN), and Convolutional Neural Network (CNN), to identify dust aerosols in the daytime satellite images from the Visible Infrared Imaging Radiometer Suite (VIIRS) under cloud-free conditions on a global scale. In order to train the ML algorithms, we collocated the state-of-the-art dust detection product from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) with the VIIRS observations along the CALIOP track. The 16 VIIRS M-band observations with the center wavelength ranging from deep blue to thermal infrared, together with solar-viewing geometries and pixel time and locations, are used as the predictor variables. Four different sets of training input data are constructed based on different combinations of VIIRS pixel and predictor variables. The validation and comparison results based on the collocated CALIOP data indicate that the FFNN method based on all available predictor variables is the best performing one among all methods. It has an averaged dust detection accuracy of about 81%, 89%, and 85% over land, ocean and whole globe, respectively, compared with collocated CALIOP. When applied to off-track VIIRS pixels, the FFNN method retrieves geographical distributions of dust that are in good agreement with on-track results as well as CALIOP statistics. For further evaluation, we compared our results based on the ML algorithms to NOAA’s Aerosol Detection Product (ADP), which is a product that classifies dust, smoke, and ash using physical-based methods. The comparison reveals both similarity and differences. Overall, this study demonstrates the great potential of ML methods for dust detection and proves that these methods can be trained on the CALIOP track and then applied to the whole granule of VIIRS granule.