Innovative Approaches to Enhance Human Respiratory Tract Representation for Radiation Dosimetry

Purpose:
Accurate representation of the human respiratory tract (HRT) anatomy is crucial for understanding the structure-function relationship in the respiratory system. Historically, this domain has been primarily driven by morphometric analyses of lung casts. These methods have been incrementally enriched employing computed tomography (CT) imaging, albeit on a limited cohort. Therefore, this study aims to enhance the HRT representation for radiation dosimetry evaluation, addressing the limitations of existing simplified models by the International Commission on Radiological Protection (ICRP); which do not include of children and senior citizens, in the event of a radiological disaster.

Methods:

Utilizing CT scans from 542 individuals across a wide age range (2 to 90 years old) population, 3D models of the HRT were reconstructed. These models were segmented using a Convolutional Neural Network (CNN), and critical physiological parameters were extracted with a Python library developed for automatic parameter extraction, encompassing trachea length, average trachea diameter, G0-to-G1 branching angle, and overall reconstruction volume. The study employs Principal Component Analysis (PCA), Random Forest algorithms, and k-clustering to analyze the extracted data.

Results:

The results revealed critical parameters such as tracheal diameter and bronchial angle, which are pivotal for predicting patient age, sex, and weight with an error margin of less than 18\%. The presented methodology efficiently condensed the extensive variability of airway morphology into nine representative models per sex group, covering the maximum population variance. This innovative approach facilitates subject-specific dose reconstruction, surpassing the capabilities of existing models.

Conclusion:

Our study provides a significant leap forward in accurately representing the HRT for radiation dosimetry purposes. By utilizing advanced machine learning techniques and 3D modeling, a framework was developed that offers a scalable solution for personalized medical physics applications. This work holds considerable interest for the medical physics community, promising a paradigm shift in personalized radiation dosimetry models.