Why AI is securing a stronger foothold in radiology
Collaboration between artificial intelligence companies and mainstream vendors boosts likelihood that the technology will be integrated into radiologists’ workflows.
It’s clear that artificial intelligence is continuing to stake territory in radiology, and professionals are looking to incorporate the technology in helping them deliver care.
For the first time at this month’s European Congress of Radiology, there was a dedicated section of the Expo for AI—the Artificial Intelligence Future Lab. There were also a handful of medical imaging AI companies dotted around the main exhibition halls, and most of the major vendors found an angle to add AI to their booths.
From walking the exhibition floor, it’s clear that AI continues to make inroads into medical imaging and the pace of technology commercialization is accelerating.
For AI to become mainstream in medical imaging, the tools need to be fully integrated into the radiologists’ existing workflow. Most generalist radiologists will prefer to access the results from AI algorithms from within their diagnostic viewer, which in most cases today is a PACS. Coming out of PACS to a dedicated AI platform adds an extra step in the process and hence additional time, which is particularly hard to justify for AI platforms with a narrow offering of algorithms. An exception could be targeted use cases, such as breast imaging, where radiologists may already be used to working with a computer-aided detection (CADe) workstation.
Radiologists who already use a dedicated advanced visualization (AV) platform may be comfortable with accessing certain AI tools, such as quantitative imaging tools, in an AV environment. However, some convergence of AV and PACS is occurring, and over time, we expect to see more AV tools made available in PACS.
As the industry transitions to enterprise imaging and universal viewers, AI developers need to forge partnerships with the leading imaging IT vendors to best position themselves for future growth. In the coming years, we expect to see PACS, AV and other clinical IT platforms, such as oncology, pathology, breast imaging, cardiology, dermatology and orthopaedics, absorbed into enterprise platforms, giving AI developers access to a richer set of data.
Forming partnerships with PACS vendors can enable tighter integration of AI tools for an enhanced user experience. The AI results can be directly overlaid on the images, and the radiologist can make edits and annotations. A tight PACS integration can also give access to the worklist so that cases can be prioritised in the reading list based on the initial AI findings.
The Dutch company Aidence is taking this approach and has integrated its Veye Chest Detection solution with PACS from Agfa and Sectra, negating the need for a specific AI user interface. Veye Chest Detection automatically detects and marks pulmonary nodules on chest CT scans.
Aidoc has taken a similar approach and fully integrates its AI solution in the PACS workflow, without a dedicated AI platform. While most AI solutions are built for specific pathologies, Aidoc takes a different approach and analyzes complete body areas to detect potential abnormalities. The software can then prioritize cases in the worklist. Aidoc is targeting acute findings only and does not provide quantification. The company has a co-marketing agreement with Agfa and is seeking partnerships with other PACS vendors.
As we mentioned in a report on last December’s RSNA conference, online AI marketplaces provide algorithm developers with workflow integration and a route to market. EnvoyAI used the European Congress of Radiology for its European launch and now has 19 developers with 46 products on its platform. Siemens also showed its Digital Ecosystem marketplace at ECR 2018. It was first announced at HIMSS17 with a handful of partners on board, and Siemens has added several new partners over the last year. The current list of partners now includes solutions for image analysis (AMRA, Arterys, Circle Cardiovascular Imaging, Combinostics, HeartFlow, Mint Medical, Pie Medical Imaging, Precision Image Analysis, SyntheticMR), surgical (ExplORer Surgical, mediCAD), teleradiology (Second Opinions, TMC) and operational and business Analytics (Cranberry Peak, Stroll Health, Viewics, Dell-EMC).
During ECR, Philips announced the launch of an AI development environment, HealthSuite Insights, which includes the Insights Marketplace, an ecosystem of AI tools from Philips and third parties (from late 2018).
We expect more modality and imaging IT vendors will launch AI marketplaces this year.
While many of the medical imaging AI startups are focused on image analysis, imaging IT and modality vendors are applying AI to a broader range of imaging applications, including pre-, intra- and post-scan.
For example, in pre-scan, AI can be used to ensure the patient is correctly positioned and the optimal scan protocol is used. During the scan, AI can detect additional abnormalities than the initial target and optimise the scan for incidental findings, potentially eliminating the need for additional scans. Post-scan, in additional to the usual post processing applications, such as detection, segmentation and quantification, AI can be applied for practice management (e.g. quality and outcome assessment tools) and radiologist workflow, such as adaptive hanging protocols, automatic retrieval of priors, etc. An additional example is the use of deep learning to improve the image quality of low-dose CT images.
Many of the modality and imaging IT vendors are initially focussing their in-house AI developments on practice management and workflow, with clinical applications seen as a longer-term play. Partnering with AI specialists is an effective way of establishing an initial offering of clinical AI solutions, so as not to be seen as a laggard. Since radiologist workflow is essentially ingrained in the PACS, it makes sense that the imaging IT vendors want to control the development of AI-powered workflow tools. Moreover, there are fewer challenges with bringing these solutions to market, such as fewer regulatory requirements.
However, overall, AI didn’t seem to generate the same buzz and excitement as it did at RSNA. Perhaps this is a positive sign that AI is now descending from the peak of the hype cycle. Or perhaps it’s a sign that European radiologists aren’t yet fully embracing AI, waiting for industry leaders to champion its use.
For the first time at this month’s European Congress of Radiology, there was a dedicated section of the Expo for AI—the Artificial Intelligence Future Lab. There were also a handful of medical imaging AI companies dotted around the main exhibition halls, and most of the major vendors found an angle to add AI to their booths.
From walking the exhibition floor, it’s clear that AI continues to make inroads into medical imaging and the pace of technology commercialization is accelerating.
For AI to become mainstream in medical imaging, the tools need to be fully integrated into the radiologists’ existing workflow. Most generalist radiologists will prefer to access the results from AI algorithms from within their diagnostic viewer, which in most cases today is a PACS. Coming out of PACS to a dedicated AI platform adds an extra step in the process and hence additional time, which is particularly hard to justify for AI platforms with a narrow offering of algorithms. An exception could be targeted use cases, such as breast imaging, where radiologists may already be used to working with a computer-aided detection (CADe) workstation.Radiologists who already use a dedicated advanced visualization (AV) platform may be comfortable with accessing certain AI tools, such as quantitative imaging tools, in an AV environment. However, some convergence of AV and PACS is occurring, and over time, we expect to see more AV tools made available in PACS.
As the industry transitions to enterprise imaging and universal viewers, AI developers need to forge partnerships with the leading imaging IT vendors to best position themselves for future growth. In the coming years, we expect to see PACS, AV and other clinical IT platforms, such as oncology, pathology, breast imaging, cardiology, dermatology and orthopaedics, absorbed into enterprise platforms, giving AI developers access to a richer set of data.
Forming partnerships with PACS vendors can enable tighter integration of AI tools for an enhanced user experience. The AI results can be directly overlaid on the images, and the radiologist can make edits and annotations. A tight PACS integration can also give access to the worklist so that cases can be prioritised in the reading list based on the initial AI findings.
The Dutch company Aidence is taking this approach and has integrated its Veye Chest Detection solution with PACS from Agfa and Sectra, negating the need for a specific AI user interface. Veye Chest Detection automatically detects and marks pulmonary nodules on chest CT scans.
Aidoc has taken a similar approach and fully integrates its AI solution in the PACS workflow, without a dedicated AI platform. While most AI solutions are built for specific pathologies, Aidoc takes a different approach and analyzes complete body areas to detect potential abnormalities. The software can then prioritize cases in the worklist. Aidoc is targeting acute findings only and does not provide quantification. The company has a co-marketing agreement with Agfa and is seeking partnerships with other PACS vendors.
As we mentioned in a report on last December’s RSNA conference, online AI marketplaces provide algorithm developers with workflow integration and a route to market. EnvoyAI used the European Congress of Radiology for its European launch and now has 19 developers with 46 products on its platform. Siemens also showed its Digital Ecosystem marketplace at ECR 2018. It was first announced at HIMSS17 with a handful of partners on board, and Siemens has added several new partners over the last year. The current list of partners now includes solutions for image analysis (AMRA, Arterys, Circle Cardiovascular Imaging, Combinostics, HeartFlow, Mint Medical, Pie Medical Imaging, Precision Image Analysis, SyntheticMR), surgical (ExplORer Surgical, mediCAD), teleradiology (Second Opinions, TMC) and operational and business Analytics (Cranberry Peak, Stroll Health, Viewics, Dell-EMC).
During ECR, Philips announced the launch of an AI development environment, HealthSuite Insights, which includes the Insights Marketplace, an ecosystem of AI tools from Philips and third parties (from late 2018).
We expect more modality and imaging IT vendors will launch AI marketplaces this year.
While many of the medical imaging AI startups are focused on image analysis, imaging IT and modality vendors are applying AI to a broader range of imaging applications, including pre-, intra- and post-scan.
For example, in pre-scan, AI can be used to ensure the patient is correctly positioned and the optimal scan protocol is used. During the scan, AI can detect additional abnormalities than the initial target and optimise the scan for incidental findings, potentially eliminating the need for additional scans. Post-scan, in additional to the usual post processing applications, such as detection, segmentation and quantification, AI can be applied for practice management (e.g. quality and outcome assessment tools) and radiologist workflow, such as adaptive hanging protocols, automatic retrieval of priors, etc. An additional example is the use of deep learning to improve the image quality of low-dose CT images.
Many of the modality and imaging IT vendors are initially focussing their in-house AI developments on practice management and workflow, with clinical applications seen as a longer-term play. Partnering with AI specialists is an effective way of establishing an initial offering of clinical AI solutions, so as not to be seen as a laggard. Since radiologist workflow is essentially ingrained in the PACS, it makes sense that the imaging IT vendors want to control the development of AI-powered workflow tools. Moreover, there are fewer challenges with bringing these solutions to market, such as fewer regulatory requirements.
However, overall, AI didn’t seem to generate the same buzz and excitement as it did at RSNA. Perhaps this is a positive sign that AI is now descending from the peak of the hype cycle. Or perhaps it’s a sign that European radiologists aren’t yet fully embracing AI, waiting for industry leaders to champion its use.
More for you
Loading data for hdm_tax_topic #better-outcomes...