Artificial Intelligence-Enabled Analysis of Thermography to Diagnose Acute Decompensated Heart Failure

Analyzing skin temperature in heart failure is an important medical practice that could assist to identify poor perfusion. Thermography, a technique that captures infrared radiation from tissues, could quantify these temperatures and thermal gradients. It has not been evaluated in patients with acute decompensated heart failure (ADHF) before.

The purpose of this study was to assess the performance of thermography in the diagnosis of ADHF.

A cross-sectional study was performed, including consecutive patients hospitalized with ADHF diagnosed by an expert heart failure team. Patients hospitalized for other cardiac disorders without ADHF were included as controls. Ten thermal photos of each patient were taken within the first 4 hours after admission in a cardiac care unit. Specific thermal spots, averages, and gradients were analyzed. Thermography's diagnostic properties for ADHF detection were evaluated using machine learning with the extreme gradient boosting model.

Sixty patients were included: 30 cases with ADHF and 30 controls. The mean age was 63.4 years (SD: 13.3), and 38 (63.3%) were males. Thermal points and averages showed lower temperature, while gradients were higher in the ADHF group, being all statistically significant between groups. The properties of the blend between thermography and artificial intelligence to detect ADHF had 84% sensitivity and 52% specificity. The area under the curve was 0.82 (95% CI: 0.73-0.91).

Thermography demonstrated differences between patients with ADHF and those with other cardiological disorders. In this proof of concept, combining thermography with artificial intelligence enabled the detection of ADHF in subjects hospitalized in a cardiac care unit.