Augmenting AI: Increasing Throughput, Quality and Validity of Imaging Data for Biomedical AI
Lead Research Organisation:
ffei (United Kingdom)
Department Name: UNLISTED
Abstract
The use of artificial intelligence (AI) in biological discovery and digital healthcare is increasing at rate. Digital imaging provides large quantities of diagnostic data in formats amenable to widespread adoption and automated analysis. As a result, research and commercial opportunities are arising to enhance and adapt current technologies to improve efficiency and automation with AI. However, in order to ensure the safest and most reliable deployment of these technologies in the digital era, there is a core requirement to ensure that the data upon which automated analyses are performed are of the highest quality and validity to ensure reliably positive outcomes. Furthermore, to warrant the maximum benefits of automation are reached, analysis devices must perform at the highest throughput and efficiency - a process that can be self-fulfilling by the integration of AI-workflow and Industry 4.0 approaches.
To augment biomedical AI the applicant proposes a Portfolio Fellowship that will enhance, integrate and optimise FFEI's past, present and future technologies in bio-imaging and digital analysis workflow. This ambitious project will develop four core technologies, each enhancing a stage of the AI imaging pipeline. Technologies will start from different stages of market-readiness to ensure commercial and grant deliverables are manageable and realised. The overlapping stages of development will lead to a sustainable balance of commercial and research activities, whilst incrementally priming AI imaging markets for the emergence of modular, end-to-end AI technology from FFEI that can provide solutions that are adaptable and integrative to most segments of the digital healthcare market. A long-term objective is to integrate all the core developments into an FFEI 'smart lab' product, in which a single, modular device can perform all essential activities of biomedical AI laboratories.
The Fellow aims to develop FFEI's biomedical capabilities by establishing a research environment to collect baseline metrics of current technology as a starting point for enhancement. The project will prototype new and established FFEI technology with flexibility to integrate emerging concepts from a network of academic partners. Ultimately, the objective will be for the Fellow's team to be able to dynamically test biological, mechanical and computational concepts to better achieve end-to-end optimisation of image data for AI. A key objective is to prove augmentation validity in refining end-to-end medical AI efficiency and reliability. To medically validate these technologies beyond concept, the Fellow will collaborate with NHS partners in parallel to FFEI productisation, allowing for iterative optimisations and case-data for accreditation. Enhancement of workflow processes with AI will require practical assessment and expert consultation, therefore the Fellow will create and lead a consortium of academic and NHS collaborators with expertise in biomedical R&D, diagnostics and AI analysis, further raising awareness through dissemination of peer-reviewed data.
An essential component of the project's success will be the creation of a new 'AI imaging' team under the Fellow's leadership. FFEI have a highly experienced and established R&D imaging team into which the Fellow will recruit new members to grow FFEI's Life Science business, to bolster this team and explore new concepts whilst learning the skills of blending innovative thinking with commercial application. In return, the new team will bring fresh talent to FFEI, with anticipated recruiting of AI software, advanced opto-mechanical and biological experts, developing a team to take FFEI into a new age of AI augmentation and commercial success. The project will benefit the Fellow with personal development opportunities in business management, team leadership and commercial collaborations, under the mentorship and support of the FFEI executive team.
To augment biomedical AI the applicant proposes a Portfolio Fellowship that will enhance, integrate and optimise FFEI's past, present and future technologies in bio-imaging and digital analysis workflow. This ambitious project will develop four core technologies, each enhancing a stage of the AI imaging pipeline. Technologies will start from different stages of market-readiness to ensure commercial and grant deliverables are manageable and realised. The overlapping stages of development will lead to a sustainable balance of commercial and research activities, whilst incrementally priming AI imaging markets for the emergence of modular, end-to-end AI technology from FFEI that can provide solutions that are adaptable and integrative to most segments of the digital healthcare market. A long-term objective is to integrate all the core developments into an FFEI 'smart lab' product, in which a single, modular device can perform all essential activities of biomedical AI laboratories.
The Fellow aims to develop FFEI's biomedical capabilities by establishing a research environment to collect baseline metrics of current technology as a starting point for enhancement. The project will prototype new and established FFEI technology with flexibility to integrate emerging concepts from a network of academic partners. Ultimately, the objective will be for the Fellow's team to be able to dynamically test biological, mechanical and computational concepts to better achieve end-to-end optimisation of image data for AI. A key objective is to prove augmentation validity in refining end-to-end medical AI efficiency and reliability. To medically validate these technologies beyond concept, the Fellow will collaborate with NHS partners in parallel to FFEI productisation, allowing for iterative optimisations and case-data for accreditation. Enhancement of workflow processes with AI will require practical assessment and expert consultation, therefore the Fellow will create and lead a consortium of academic and NHS collaborators with expertise in biomedical R&D, diagnostics and AI analysis, further raising awareness through dissemination of peer-reviewed data.
An essential component of the project's success will be the creation of a new 'AI imaging' team under the Fellow's leadership. FFEI have a highly experienced and established R&D imaging team into which the Fellow will recruit new members to grow FFEI's Life Science business, to bolster this team and explore new concepts whilst learning the skills of blending innovative thinking with commercial application. In return, the new team will bring fresh talent to FFEI, with anticipated recruiting of AI software, advanced opto-mechanical and biological experts, developing a team to take FFEI into a new age of AI augmentation and commercial success. The project will benefit the Fellow with personal development opportunities in business management, team leadership and commercial collaborations, under the mentorship and support of the FFEI executive team.
Publications
Romanchikova M
(2022)
The need for measurement science in digital pathology.
in Journal of pathology informatics
| Description | - Building a Team - recruited a PhD histology expert, secured internal resources, formed a dedicated QA in AI research group. Interweaves with experienced FFEI resource for continual knowledge transfer and preparing for future growth of the technology. - Software toolkit - proprietary image format variation is a major blocker for pathology AI interoperability. Via collaboration and internal expertise, we expanded direct file format compatibility and developed workarounds for others, expanding AI market reach. In doing so, new areas of image QA have been identified - more images from more manufacturers of varying quality creates challenges to AI that must be first detected. An SDK was developed to integrate findings into AI workflow. - R&D lab - lab capabilities were upgraded to expand the range and throughput of tested histology variables in our calibration target. We can now quickly adapt our technology to be responsive to specific pathology AI applications. - Applications in QA - a global network of clinicians and prostate AI developers established the validation and impact of physically calibrating pathology scanner output to the ground-truth, revealing increased accuracy versus native images and increased validation against human assessment. Significantly, this also demonstrated that less data is needed for increased reliability using physical calibration rather than arbitrary normalisation by AI methods like cycleGANs and Macenko. - Raising awareness of technology impact - presented Fellowship data in the form of conference speaking and/or scientific poster at x5 international events (2x UK, 2x USA, 1x Switzerland) and x2 big pharma interdepartmental virtual seminars - this led to engagement with several academic and industrial AI collaborations. - Ongoing Collaboration - three collaborations, each approaching a different sector of the biomedical AI market, were officially established and are close to fruition of producing papers and moving out of R&D into pre-market validation for commercial impact. Further collaborations resulting from Yr 1 public engagement activities are due to begin and are likely to expand in 2023. |
| Exploitation Route | - The application of colour calibration metadata to pathology images is FTO, with open-source available (some by FFEI) and well established in many digital markets - therefore the methodology used for colour correction in the AI projects of the grant is universal after the data is generated from proprietary FFEI technology. How individual AI algorithms use that calibrated data is proprietary to specific developers, but the creation of QA, calibrated image data can and should be used by all. - FFEI are commercialising other research findings, but intend into products adoptable by non-colour experts that provide simplified analysis for clear understanding and will be non-exclusive - i.e. impact is obtainable by the whole market, not monopolised by one organization. - FFEI will continue to disseminate scientific findings in this field to raise awareness of the topic and then direct to the resources needed for solutions. Such resources and industrial standards are likely to be developed by the national and international working groups and committees the Fellow and his team now sit on as part of this grant. |
| Sectors | Digital/Communication/Information Technologies (including Software),Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology |
| Description | Demonstration that scanner-agnostic physical calibration of device image colour for AI, though in an academic setting, provides a quick route to increased accuracy and validation of developing AI, whilst requiring less image data to achieve industry expectations of performance. This provides opportunities for rapid commercialisation of AI with independent validation for regulatory acceptance and drives the market segment to demand higher quality data from scanner and data repositories to provide better ROI by getting to market faster and more competitively. In turn, this means patients will benefit from the efficiency and throughput actions of wider array of deployable AI. |
| First Year Of Impact | 2022 |
| Sector | Digital/Communication/Information Technologies (including Software),Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology |
| Impact Types | Societal,Economic,Policy & public services |
| Description | UKRI Data to Early Diagnostics Interoperability Workstream |
| Geographic Reach | National |
| Policy Influence Type | Contribution to a national consultation/review |
| Impact | Identified key challenges remaining for complete interoperability in digital medicine. Made meaningful progress where possible in predetermined areas. Documented challenges that have been overcome and established resources developed or deployed across the network. Future work packages extended to include NHS Digital and IHE-UK. |
| Description | Colour Validation of Digital Workflow in Pathology |
| Organisation | Barco |
| Country | Belgium |
| Sector | Private |
| PI Contribution | Calibration technology, expertise and analysis by FFEI's skilled team (4 people involved) for the correction of clinical digital pathology scanner digital colour output when displayed on medical-grade monitors for human assessment - testing essential role as part of AI validation. Data processed from clinical settings using multiple scanner vendors. Planning for white paper/scientific publication. |
| Collaborator Contribution | Proof of concept (3 people involved) that FFEI colour technology and profiles can be used independently with a clinical-grade monitor to standardise and quantify image quality for pre-AI validation and clinical application. Access to cross-industry scanners with clinical users in a real-world, independent deployment. Planning for white paper/scientific publication. |
| Impact | Testing of scope when FFEI technology is applied to historic clinical images with or without scanner calibration. Demonstration of clear visual standardisation to a non-scanner colour expert when technology is independently assessed - indicates AI using FFEI technology can be reliably validated by humans using clinical-grade monitors. Identification of pathology-scanner imaging quality issues (i.e. pre-FFEI technology) that FFEI may need to adapt for in future QC investigations. Multi-disciplinary collaboration - optomechanical and digital colour experts from industry, biochemists and histology scientists for ground-truth colour reproducibility and assessment, leasdership and coordination of international study, display monitor experts in human validation of findings. |
| Start Year | 2022 |
| Description | PACMAN (Prostate AI with Color Managment and Normalisation) project |
| Organisation | Karolinska Institute |
| Country | Sweden |
| Sector | Academic/University |
| PI Contribution | Calibration technology development, activity and expertise for the analysis and standardisation of colour from multi-source pathology scanners across several Scandinavian countries and hospitals to provide a measure of accuracy and digital control over data quality used by prostate cancer AI. Contribution to pre-AI analysis of data and preparation of scientific manuscript. Mutual dissemination of research findings on the world-stage via conference speaking roles. |
| Collaborator Contribution | Access to multi-source pathology scanners and pathologists across several Scandinavian countries and hospitals to generate industry-wide metrics, statistical analysis and software tools to standardise colour output. Prostate AI expertise and analysis of data and preparation of scientific manuscript. Mutual dissemination of research findings on the world-stage via conference speaking roles. |
| Impact | World-first findings on the improved control, accuracy and reliability of an independent AI algorithm using a physical method of calibrating and normalising the digital colour output of pathology scanners used in cancer research from a variety of leading vendors in the market. Multi-disciplinary collaboration - optomechanical and digital colour experts from industry, biochemists and histology scientists for ground-truth colour reproducibility and assessment, academic development and re-training of AI, PI coordination of international study, medical pathologists in human validation of AI findings. |
| Start Year | 2022 |
| Description | IEC TC 62/PT63450 - Assessment of AI components of medical devices |
| Form Of Engagement Activity | A formal working group, expert panel or dialogue |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Policymakers/politicians |
| Results and Impact | Aim of IEC TS62 is to prepare international standards, and other publications, with focus on safety and performance of medical equipment, software, and systems. TC 62 coordinates and addresses emerging technologies with a leading focus on defining assessment and validation of AI-driven components (hardware or software) used in medical imaging systems. My colleague (time funded by the grant) represents the UK in this international think-tank and is driving the topic and policy of colour calibration as per the grant's remit - the output may take several years but will generate international standards such as used by ISO, FDA and MHRA, with the grant technology at the forefront. |
| Year(s) Of Engagement Activity | 2022,2023 |
| URL | https://www.iec.ch/dyn/www/f?p=103:14:709479161484870::::FSP_ORG_ID:28783 |
| Description | IHE-UK Digital Pathology SIG |
| Form Of Engagement Activity | A formal working group, expert panel or dialogue |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Professional Practitioners |
| Results and Impact | The creation of this group was championed by NHS Digital, National Physical Laboratory, British Standards Institute, Digital Pathology Centres of Excellence and industry suppliers. Focus is on Interoperability standards and frameworks for pathology data (in particular structured data), Terminologies and codification for pathology and Digital imaging and open standards in pathology. The last topic is of particular relevance to the funding, as establishment of FFEI colour standards a prerequisite for AI is key to research deployment and commercial scope. Leading outcome so far is the interaction with industry, NHS and academic leaders who can influence the metrology-based advantages of the emerging grant technology. |
| Year(s) Of Engagement Activity | 2022,2023 |