A single inhalation CT scan combined with a deep learning model can accurately diagnose and stage COPD.
RT’s Three Key Takeaways:
- Single CT Scan Sufficiency: A deep learning model can diagnose and stage COPD accurately using just one inhalation CT scan, eliminating the need for expiratory imaging.
- Increased Accessibility: This approach reduces patient costs, discomfort, and radiation exposure, making COPD diagnostics more accessible, especially in facilities without dual-phase CT capabilities, according to researchers.
- Enhanced Accuracy with Data: Combining clinical data with the deep learning model improves diagnostic precision, matching results from more complex imaging protocols.
Using just one inhalation lung CT scan, a deep learning model can accurately diagnose and stage chronic obstructive pulmonary disease (COPD), according to a study published in Radiology: Cardiothoracic Imaging.
A spirometry test is traditionally used to diagnose COPD. It measures lung function through the quantity of air that can be inhaled and exhaled as well as the speed of exhalation. CT images of the lungs can aid in COPD diagnosis. The procedure typically requires two image acquisitions, one at full inhalation and one at normal exhalation.
“Although studies have recently shown that lung structure, quantitatively measured using lung CT, can supplement COPD severity staging, diagnosis, and prognosis, many of these studies require the acquisition of two CT images,” says study author Kyle A. Hasenstab, PhD, assistant professor of Statistics and Data Science at San Diego State University, California, in a release. “However, this type of protocol is not clinically standard across institutions.”
Challenges with Dual-Phase CT Imaging
Some hospitals are unable to implement expiratory CT protocols due to the added training requirements.
“Implementation of expiratory CT protocols may not be feasible at many institutions due to the need for technologist training to acquire the images and radiologist training to interpret the images,” says Hasenstab in a release.
Additionally, some elderly patients with impaired lung function struggle with holding their breath, as is required during exhalation image acquisition. This may impact the quality of CT images and the accuracy of diagnosis.
A Single CT Scan and CNN Model
Hasenstab and colleagues hypothesized that a single inhalation CT acquisition combined with a convolutional neural network (CNN) and clinical data would be sufficient for COPD diagnosis and staging. A CNN is a type of artificial neural network that uses deep learning to analyze and classify images.
In this retrospective study, the inhalation and exhalation lung CT images and spirometry data were acquired from 8,893 patients from November 2007 to April 2011. The average age of the patients included in the study was 59 years, and all had a history of smoking.
The CNN was trained to predict spirometry measurements using clinical data and either a single-phase or multi-phase lung CT.
The spirometry predictions were then used to predict the Global Initiative for Obstruct Lung Disease (GOLD) stage. The GOLD system classifies the severity of a patient’s COPD into one of four stages, with one classified as mild COPD and four classified as very severe COPD.
Study Results and Implications
The results of the study showed that a CNN model developed using only a single respiratory phase CT image accurately diagnosed COPD and was also accurate within one GOLD stage. The model performed similarly to COPD diagnoses that used combined inhalation and exhalation CT measurements.
“Although many imaging protocols for COPD diagnosis and staging require two CT acquisitions, our study shows that COPD diagnosis and staging is feasible with a single CT acquisition and relevant clinical data,” says Hasenstab in a release.
When clinical data was added, the CNN model’s predictions were even more accurate.
CNN models that used only inhalation or exhalation data respectively performed the same. This suggests that certain markers used for COPD diagnosis may overlap across images.
“Reduction to a single inspiratory CT acquisition can increase accessibility to this diagnostic approach while reducing patient cost, discomfort, and exposure to ionizing radiation,” says Hasenstab in a release.
Photo caption: Visualization of attention maps from the residual attention networks for the inspiratory convolutional neural network and expiratory convolutional neural network models. Attention maps highlight regions of the input image contributing to the network’s predictions. Low-level features (minor details, such as texture) and high-level features (major details, such as shape) are displayed overlaying a single coronal section of the original input non-contrast CT images. Overlay color indicates strength of attention (ie, strength of contribution) ranging from low attention (blue) to high attention (dark red). One example for each Global Initiative for Chronic Obstructive Lung Disease stage is shown.
Photo credit: Radiological Society of North America
