Explainable deep learning predicts pulmonary blood flow from x-rays

Deep learning can help physicians get a more accurate measurement of pulmonary systemic blood flow in patients with congenital heart disease, reported authors of a Jan. 22 study published in JAMA Cardiology.

The artificial intelligence platform predicted pulmonary to systemic flow ratio by analyzing more than 1,000 chest radiographs, beating out clinicians with varying levels of experience. Lead researcher Shuhei Toba, MD, and colleagues believe their tool can illuminate information invisible to clinicians, and perhaps improve their clinical decision-making.

“Because of our model’s capability to quantitatively predict the pulmonary to systemic flow ratio from chest radiographs and to outperform clinicians, the present proof-of-concept study suggests that there may be hidden information in routine imaging tests that deep learning can identify, adding clinical value,” Toba, with Mie University Graduate School of Medicine in Japan, and colleagues added.

Getting a precise measurement of pulmonary to systemic blood flow is important in assessing patients with congenital heart disease, but doing so typically requires invasive cardiac catheterization. Echocardiography and MRI have proven useful for predicting this measurement, but both have a number of limitations. And chest x-ray is less invasive, but more subjective and qualitative.

Hypothesizing that deep learning could help, Toba et al. retrospectively tested their algorithm in 657 patients who underwent a total of 1,031 cardiac catheterizations. They also compared it to pediatric cardiology experts in a subgroup of 78 random patients who underwent 100 total catheterizations.

The deep learning algorithm correctly classified 64 of 100 chest x-rays, beating out the experts' 49 out of 100 accurate readings.

Additionally, and unlike many deep learning platforms, the authors were able to gain some insight into how the algorithm came to its conclusions. Employing three different methods to visualize the reasoning, Toba and colleagues found their model recognized and focused on structures in the lung fields and the area around the heart for its predictions. These regions are different from expert consensus contained in standard textbooks, the authors noted.

The team did acknowledge that their visualization methods only revealed limited parts of their deep learning model and therefore may not fully detail the reasoning behind the blood flow predictions.

“Further studies are warranted to improve the performance of the model and to understand how the model predicts the pulmonary to systemic flow ratio,” Toba and co-investigators wrote.

""

Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.

Around the web

CMS has more than doubled the CCTA payment rate from $175 to $357.13. The move, expected to have a significant impact on the utilization of cardiac CT, received immediate praise from imaging specialists.

The newly cleared offering, AutoChamber, was designed with opportunistic screening in mind. It can evaluate many different kinds of CT images, including those originally gathered to screen patients for lung cancer. 

AI-enabled coronary plaque assessments deliver significant value, according to late-breaking data presented at TCT. These AI platforms have gained considerable momentum in recent months, receiving expanded Medicare coverage in addition to a new Category I CPT code.

Trimed Popup
Trimed Popup