RESUMO
OBJECTIVES: How do providers of artificial intelligence (AI) solutions propose and legitimize the values of their solutions for supporting diagnostic radiology workflow? METHODS: We systematically analyze 393 AI applications developed for supporting diagnostic radiology workflow. We collected qualitative and quantitative data by analyzing around 1250 pages of documents retrieved from companies' websites and legal documents. Five investigators read and interpreted collected data, extracted the features and functionalities of the AI applications, and finally entered them into an excel file for identifying the patterns. RESULTS: Over the last 2 years, we see an increase in the number of AI applications (43%) and number of companies offering them (34%), as well as their average age (45%). Companies claim various value propositions related to increasing the "efficiency" of radiology work (18%)-e.g., via reducing the time and cost of performing tasks and reducing the work pressure-and "quality" of offering medical services (31%)-e.g., via enhancing the quality of clinical decisions and enhancing the quality of patient care, or both of them (28%). To legitimize and support their value propositions, the companies use multiple strategies simultaneously, particularly by seeking legal approvals (72%), promoting their partnership with medical and academic institutions (75%), highlighting the expertise of their teams (56%), and showcasing examples of implementing their solutions in practice (53%). CONCLUSIONS: Although providers of AI applications claim a wide range of value propositions, they often provide limited evidence to show how their solutions deliver such systematic values in clinical practice. KEY POINTS: ⢠AI applications in radiology continue to grow in number and diversity. ⢠Companies offering AI applications claim various value propositions and use multiple ways to legitimize these propositions. ⢠Systematic scientific evidence showing the actual effectiveness of AI applications in clinical context is limited.
Assuntos
Inteligência Artificial , Radiologia , Humanos , Fluxo de Trabalho , Radiografia , Coleta de DadosRESUMO
AIMS: First, to establish the diagnostic performance of the pressure gradient at a standardised mean velocity (dPv) as derived from the cycle-averaged stenosis pressure gradient-velocity (dP-v) relationship obtained by administration of adenosine and, second, to determine whether dPv can be assessed from contrast medium-induced submaximal hyperaemia. METHODS AND RESULTS: Distal coronary pressure and velocity were simultaneously recorded in 64 patients during the response to intracoronary injection of adenosine. dPv was assessed at velocities between 20 and 50 cm/s. The pressure gradient at a mean flow velocity of 30 cm/s (dPv30) yielded an excellent diagnostic performance against FFR ≤0.8 (area under the curve 0.96; sensitivity 84%; specificity 96%; accuracy 89%). In a subgroup of 21 patients, measurements were repeated throughout contrast medium-induced reactive hyperaemia. Peak velocity and pressure gradient were lower compared to adenosine, but the course of the corresponding dP-v relationships coincided very well over the common velocity range, with no difference in dPv30. CONCLUSIONS: dPv30 reliably detects functionally significant coronary lesions. It derives from stenosis haemodynamics and can be obtained with submaximal hyperaemia, such as following injection of contrast medium, thereby obviating the maximal vasodilation by adenosine required for FFR or other established hyperaemic parameters of functional stenosis severity.