8275
Blog
Aidoc Staff

Not All AI Is Equal: A Comparative Study of Three Imaging Algorithms

The adoption of artificial intelligence (AI) in medical imaging has experienced substantial growth in recent years, with more than 450 AI-based medical devices gaining approval in both Europe and the USA. As hospitals grapple with rising costs and staff shortages, AI has emerged as a pivotal tool in making healthcare smarter and more efficient. However, in a landscape flooded with AI options and patient outcomes at stake, the significance of selecting the right AI solution is critical.

Consequently, the medical imaging community was rightfully intrigued when the Radiology Department from Stockholm South Emergency Hospital (Södersjukhuset, SÖS) unveiled findings at Rӧntgenveckan from a comparative analysis of three well-known, commercially available AI algorithms for the notification and triage of intracranial hemorrhage (ICH). While one of the algorithms is Aidoc’s, the identities of the other two providers remain respectfully confidential. The soon-to-be-published study revealed significant disparities in the performance of these algorithms, despite analysis of the same data. Aidoc’s algorithm demonstrated a 30% higher positive predictive value (PPV) and a 1.25% greater negative predictive value (NPV) when compared to the average results of the other two algorithms.

The study involved the retrospective analysis of 3902 non-contrast CT brain scans by Aidoc’s ICH algorithm run on its award-winning AI operating system (aiOS™), and two other algorithms from vendors “A” and “B” deployed via a AI marketplace PACS vendor. To ensure a fair comparison, each vendor’s algorithm complied with the same methods for the analysis of retrospective data, which meant that only DICOM-related data from the PACS was analysed, without additional support from any other data sources. Comparative sensitivity, specificity, PPV and NPV results for all three algorithms are shown below: 

SensitivitySpecificityPPVNPV
Vendor A60.2%97.1%48.4%98.2%
Vendor B62.8%97.4%52.4%98.3%
Aidoc90.3%99.0%80.3%99.5%

For full details of this study, please download the pre-publication report:

Explore the Latest AI Insights, Trends and Research

Aidoc Staff