Incidental pulmonary embolism (iPE) is a common complication in cancer patients and is a key cause of morbidity and mortality. Despite regular CT scans, studies show that cancer-associated iPE is often unreported, resulting in an increased risk of progression and new pulmonary emboli. Given the rising demand for medical imaging in European health systems, how can AI help radiologists manage incidental pathologies such as iPE and ensure patients get timely treatment?
Region Halland in Sweden is at the forefront of a pioneering effort to leverage artificial intelligence (AI) to reduce the number of incidental pathologies that are slipping through the cracks. Led by Dr. Peder Wiklund, the Department of Radiology has been exploring how they can use AI to flag suspected positive cases of incidental pathologies, reduce reporting time and deliver faster patient treatment. This article summarises the outcomes of a series of four retrospective studies using Aidoc’s AI algorithms for the notification and triage of iPE and other incidental pathologies including vertebral compression fractures (VCF).
AI can increase notification of suspected positive cases of iPE
Region Halland’s first study, published in European Radiology in 2022 assessed the potential for Aidoc’s iPE algorithm to flag suspected positive cases of iPE in CT scans of cancer patients. The study featured a retrospective analysis of 1,892 cancer patients who had received CT scans between 2018 and 2019. In this group of patients, there were 75 positive cases of iPE (4% prevalence), but only 21% of these cases were ever reported. In comparison, Aidoc’s iPE algorithm correctly flagged 91% of the iPEs with only three false positives (sensitivity 90.7%, specificity 99.8%, PPV 95.6%, NPV 99.6%), demonstrating its potential to aid radiologists by notifying them of suspected positive cases of iPE in cancer patients that would otherwise be missed.
The risk of not reporting iPE is progression or development of new iPEs
The second study from Dr. Wiklund’s group, published in Thrombosis Research earlier this year took the work a step further and showed the risk to patients of unreported cancer-associated iPE. This retrospective study featured a group of 2,960 cancer patients who had received CT scans between 2014 and 2019. Similar to the previous study, the analysis revealed that only 28% of iPE cases were ever reported. Alarmingly, over 23% of patients with overlooked iPE experienced progression of the initial PE, or developed a new one. Interestingly, patients with unreported subsegmental iPEs with multiple vessel involvement were associated with similar risk levels of iPE recurrence rates as patients with overlooked lobar/proximal or segmental iPEs. Current guidelines suggest that cancer patients with iPE benefit from treatment. Therefore, using AI to notify of suspected positive cases of iPE holds the promise of improving long-term patient outcomes.
AI can reduce the reporting time and time to treatment for iPE patients
With cancer-associated iPE, there is often a delay in reporting the finding and a delay between the finalised report and time to treatment. The third study by Dr. Wiklund’s team, recently published in Radiology: Artificial Intelligence evaluated the performance of Aidoc’s iPE AI algorithm on the report turnaround time and time to treatment for patients with cancer-associated iPE. In this retrospective study, adult cancer patients were included either before or after the implementation of Aidoc’s AI. The results showed that rates of reported iPE were significantly higher in the period after AI implementation (2.5% vs 0.8%). Furthermore, the report turnaround time decreased from 24.68 hours to 0.66 hours, and the time-to-treatment decreased from 28.05 hours to 0.98 hours after AI implementation. In conclusion, the use of AI for the notification and triage of iPE in clinical practice resulted in increased notification of suspected positive cases and significantly shorter report turnaround time and time to treatment for patients.
AI increases notification of vertebral compression fractures
Shifting gears to other incidental pathologies, this year, Dr. Wiklund’s team delivered an award-winning presentation at Rӧntgenveckan 2023 where they measured the underdiagnosis and undertreatment of vertebral compression fractures (VCF) as part of a trial using Aidoc’s UKCA and FDA-approved VCF AI algorithm1.
One in three women and one in five men over the age of 50 will suffer an osteoporotic fracture, such as a VCF during their lifetime, yet diagnosis can be a challenge because many VCFs are clinically silent or the back pain is attributed to ageing. However, many of the patients receive diagnostic scans for other reasons, which allows VCFs to be incidentally detected.
The retrospective study featured a group of 1,105 patient CT scans, exposing a 17% prevalence of VCF, with an astonishing 51.6% miss rate in new VCF patients. Of the patients with reported VCF, only 26% were subjected to clinical management.
This research project demonstrates the potential impact AI can have in triaging and managing suspected VCF. The study revealed a significant prevalence of overlooked cases and undertreatment that can be impacted by AI, offering a promising avenue for further research and implementation in healthcare settings.
The future of AI in detecting incidental pathologies
The AI initiatives showcased above emphasise the transformative potential for AI in addressing underdiagnosis and undertreatment of patients with incidental pathologies like iPE and VCF. Dr. Wiklund’s commitment to assessing AI’s role exemplifies Region Halland’s dedication to using technology to deliver high-quality patient care. This showcases that, collectively, AI and radiologists can work together to improve patient care in the face of challenging healthcare environments.
Wiklund et. al “Measuring the underdiagnosis and undertreatment of Vertebral Compression Fractures with the use of an AI algorithm”, Rӧntgenveckan 2023 ↩︎
The post Reporting the Unreported with Clinical AI: Lessons From Region Halland appeared first on Healthcare AI | Aidoc Always-on AI.
Leave a Reply