Claudia Gulli

BackgroundTo evaluate the diagnostic performance of deep learning (DL)-based fractional flow reserve derived from computed tomography (FFR-CT) applied to photon-counting coronary CT angiography (CTA) for identifying patients with hemodynamically uncertain coronary stenoses who subsequently underwent coronary revascularization.MethodsThis single-center retrospective study included 128 patients (75% male) classified as 3 (51.6%) or 4A (48.4%) according to the coronary artery disease–reporting and data system (CAD-RADS), who underwent both photon-counting coronary CTA and invasive coronary angiography (ICA). DL–based FFR-CT was computed from CTA images, with values ≤0.80 considered hemodynamically significant. Diagnostic performance for predicting downstream coronary revascularization was evaluated on per-patient and per-vessel bases using percutaneous coronary intervention (PCI) as the clinical reference endpoint. Associations between FFR-CT findings and PCI were assessed using Fisher’s exact and chi-square tests, while agreement was evaluated using Cohen’s kappa coefficient (k).ResultsPCI was performed in 80 patients (62.5%) and 98 vessels (25.5%). Per-patient sensitivity, specificity, positive predictive value, and negative predictive value were 95.0%, 93.8%, 96.2%, and 91.8%, respectively. Corresponding per-vessel values were 91.8%, 95.5%, 87.4%, and 97.2%. FFR-CT findings were significantly associated with revascularization at both levels (all p<0.001), with strong associations (contingency coefficients, 0.661 and 0.651) and excellent agreement with PCI decisions (k = 0.883 and 0.858).ConclusionDL–based FFR-CT derived from photon-counting coronary CTA demonstrated high diagnostic performance for predicting revascularization in CAD-RADS 3 and 4A patients. The integration of this algorithm into clinical workflow may refine patient selection for ICA versus conservative management, potentially reducing non-therapeutic invasive procedures.

Objectives: To assess the diagnostic accuracy of a deep learning (DL)-based algorithm for non-invasive computation of fractional flow reserve (FFR-CT) from coronary computed tomography angiography (CCTA) and to evaluate the model’s ability to automatically assign cardiovascular risk categories according to the Coronary Artery Disease–Reporting and Data System (CAD-RADS). Materials and Methods: Sixty patients with suspected coronary artery disease who underwent both CCTA and invasive coronary angiography (ICA) were retrospectively included in this multicenter study. Curved multiplanar reconstructions derived from CCTA were analyzed by the deep learning-based model to estimate FFR-CT values and to automatically assign CAD-RADS risk categories. The diagnostic performance of the software for the identification of hemodynamically significant coronary stenoses was evaluated using ICA as the reference standard. Receiver operating characteristic (ROC) curve analysis was performed to determine the area under the curve (AUC), sensitivity, and specificity on both a per-patient and per-vessel basis. Finally, agreement between CAD-RADS risk categories assigned by the DL algorithm and those determined by an expert radiologist was assessed. Results: FFR-CT demonstrated high diagnostic accuracy, with AUC of 0.935, sensitivity of 93.2%, specificity of 93.7%, and excellent agreement with reference standard (k = 0.836) on a per-patient level. Per-vessel diagnostic performance was consistently high across all major coronary arteries, with the left anterior descending artery (LAD) showing the highest accuracy (AUC = 0.932). Automated CAD-RADS classifications generated by the software showed good agreement with those assigned by human (k = 0.765). Conclusions: The DL-based model demonstrated high diagnostic accuracy and represents a promising noninvasive approach for ischemia assessment and cardiovascular risk stratification.

Artificial intelligence (AI) is reshaping radiological practice, with recent advancements in natural language processing (NLP), large language models (LLMs), and chatbot technologies opening new avenues for clinical integration. These AI-driven conversational agents have demonstrated potential in streamlining patient triage, optimizing imaging protocol selection, supporting image interpretation, automating radiology report generation, and improving communication among radiologists, referring physicians, and patients. Emerging evidence also highlights their role in decision-making, clinical data extraction, and structured reporting. While the clinical adoption of chatbots remains limited by concerns related to data privacy, model robustness, and ethical oversight, ongoing developments and regulatory efforts are paving the way for responsible implementation. This review provides a critical overview of the current and emerging applications of chatbots in radiology, evaluating their capabilities, limitations, and future directions for clinical and research integration.