Deep Learning-Based Phenotyping Of Forefoot Morphology In Hereditary Thoracic Aortic Diseases

Hereditary thoracic aortic diseases (HTAD) are often associated with multifaceted phenotypic manifestations in different anatomical districts, including skeletal abnormalities. Therefore, diagnostic criteria account for multiple parameters to compute a systemic risk score. Despite the forefoot is known to be different in HTAD, its complex morphology is difficult to be quantified objectively and it is not currently considered in diagnostic criteria. Here, we investigated the potential application of artificial intelligence to compute a HTAD risk score from smartphone-acquired images of the forefoot. To this end, we conducted a pilot study including 44 adults, of which 22 had high risk of HTAD (in line with EACTS/STS guidelines 2024 and revised Ghent criteria). The remaining 22 individuals did not show characteristic features indicative of HTAD. A deep learning architecture was then trained to compute a risk score using specific strategies to account for limited sample sizes: transfer learning and leave-one-out cross validation. The computed risk score was significantly higher in the HTAD group with respect to the control group (p < 0.0001), achieving remarkable sensitivity (82%) and specificity (91%), with an AuC of 0.94. Altogether this study highlights the usefulness of AI to assist the analysis of complex morphological traits, potentially enabling a greater number of healthcare professionals to identify patients at risk of HTAD and readily address them to a proper clinical examination. The study was approved by the Swiss Cantonal Ethics Committee with protocol number 2023-00643.