Periodic Reporting for period 1 - AICE (AI supported picture analysis in large bowel camera capsule endoscopy)
Período documentado: 2022-09-01 hasta 2024-02-29
Millions of Europeans undergo optical colonoscopy (OC) every year as a part of diagnostic procedures for bowel issues such as possible bowel cancer. For citizens and patients OC is associated with discomfort, possible complications and sick-days from work, which affects the acceptability to undergo the procedure. OC also constitutes a heavy burden on European hospital capacities.
Colon capsule endoscopy (CCE) is a technology that has the potential to replace a substantial proportion of OCs. Studies have shown that CCE is preferred by patients over OC, has a lower complication rate and can be performed out of hospital, in the home of the patient. Thus, CCE holds great potential for both patients and hospitals.
However, the current diagnostic process of CCE includes a time-consuming manual reading of the images captured by the capsule (the so-called "camera pill") done by trained medical staff, which is both expensive and prone to human error. For CCE to be a viable alternative to OC in clinical practice, these challenges need to be addressed.
Therefore, the goal of the AICE project is to develop and validate a set of artificial intelligence algorithms that can assist in the reading of the CCE images to ensure both high quality diagnostics and save crucial clinical resources.
The AICE project aims to create a complete and validated AI-assisted pathway that improves CCE diagnostics making the technology clinically viable for the good of patients, health care systems and society.
A number of the partners in AICE have been collaborating for a number of years and have completed development of several AI algorithms (AIA) for CCE diagnostics that are now in need of external clinical validation. More algorithms will be completed and prepared for validation within the first 2 years of the AICE project.
The AICE concept will focus on:
1) completing development of the remaining AIAs,
2) external validation of all of the AICE AIAs,
3) creating a clinical support system for data handling, storage and transmission,
4) developing a diagnostic pathway that considers quality, efficiency, patient preferences, ethics and economy
5) promotes the integration of AICE solutions into clinical practice via implementation guidelines and upscaling adjustments.
To achieve these goals, AICE will use an unprecedented large and diverse collection of existing patient data from nation-wide clinical studies, and will include extensive initiatives in the fields of ethics, communication and patient engagement. To ensure the right competences are present, AICE brings together clinical researchers, epidemiologists, data scientists, digital health experts, health economists, ethics researchers, SMEs, communication experts and experts in regulatory affairs.
Colon capsule endoscopy (CCE) is a technology that has the potential to replace a substantial proportion of OCs. Studies have shown that CCE is preferred by patients over OC, has a lower complication rate and can be performed out of hospital, in the home of the patient. Thus, CCE holds great potential for both patients and hospitals.
However, the current diagnostic process of CCE includes a time-consuming manual reading of the images captured by the capsule (the so-called "camera pill") done by trained medical staff, which is both expensive and prone to human error. For CCE to be a viable alternative to OC in clinical practice, these challenges need to be addressed.
Therefore, the goal of the AICE project is to develop and validate a set of artificial intelligence algorithms that can assist in the reading of the CCE images to ensure both high quality diagnostics and save crucial clinical resources.
The AICE project aims to create a complete and validated AI-assisted pathway that improves CCE diagnostics making the technology clinically viable for the good of patients, health care systems and society.
A number of the partners in AICE have been collaborating for a number of years and have completed development of several AI algorithms (AIA) for CCE diagnostics that are now in need of external clinical validation. More algorithms will be completed and prepared for validation within the first 2 years of the AICE project.
The AICE concept will focus on:
1) completing development of the remaining AIAs,
2) external validation of all of the AICE AIAs,
3) creating a clinical support system for data handling, storage and transmission,
4) developing a diagnostic pathway that considers quality, efficiency, patient preferences, ethics and economy
5) promotes the integration of AICE solutions into clinical practice via implementation guidelines and upscaling adjustments.
To achieve these goals, AICE will use an unprecedented large and diverse collection of existing patient data from nation-wide clinical studies, and will include extensive initiatives in the fields of ethics, communication and patient engagement. To ensure the right competences are present, AICE brings together clinical researchers, epidemiologists, data scientists, digital health experts, health economists, ethics researchers, SMEs, communication experts and experts in regulatory affairs.
In the first reporting period, the main focus has been on the development of the AICE algorithms (AIA) among the technical partners in the project to get ready for external technical validation and clinical validation. One scientific paper has been published around the AIA for polyp detection and segmentation, with another paper currently under review.
Alongside this, we have worked hard in the process of getting access to Scottish clinical data for the external validation of the algorithms. As is often the case, this process is more time and resource consuming than hoped, no less because the Scottish capsule data is spread across the different Scottish health boards, each of whom has a separate process for data access. This process has been lead by the Scottish partner NHS Highland in close cooperation with Coordinator RSYD to ensure that the data is compatible with the Danish data ensuring that we can validate the algorithms using this data. At the end of the first reporting period, several iterations of filling out and answering inquiries about process documents have been carried out. It is foreseen that access to Scottish data will be in place in the second half of 2024, with plans for technical and clinical AICE team members to travel to Scotland to validate the AIAs locally, as it will not be possible to merge the Scottish data into a Danish or international research database.
Along with the practical tasks of getting access to data and completing lab tests and internal validation of the AICE algorithms, we have started the process of setting up of a clinical support system for how an AI-supported CCE service could be put into clinical practice. Analyses and workshops have led to identification of a "service data flow" diagram based on the current clinical practice and IT infrastructure at Odense University Hospital. In reporting period two and three these workflows will be further developed and fleshed out, working towards a final, recommended technical diagram for implementation into clinical practice, that takes into account the clinical reality and the use of AI.
A review study has been conducted to prepare the work on the diagnostic pathway that will be a main focus in reporting period two. As part of this work, patient and end-user representatives will be invited to contribute.
Preparations for the work on a 'CCE patient support mobile application' for patients undergoing a CCE procedure has also been started in Scotland, with plans for close patient inclusion in the design, development and test phases of the process.
This work will be carried out in close collaboration with the work on developing clinical implementation guidelines. The vision is to develop the guidelines as a set of living guidelines integrated in the app, collecting in one place all the scientific knowledge on both clinical CCE procedures and on AI-supported CCE used in clinical practice.
Alongside this, we have worked hard in the process of getting access to Scottish clinical data for the external validation of the algorithms. As is often the case, this process is more time and resource consuming than hoped, no less because the Scottish capsule data is spread across the different Scottish health boards, each of whom has a separate process for data access. This process has been lead by the Scottish partner NHS Highland in close cooperation with Coordinator RSYD to ensure that the data is compatible with the Danish data ensuring that we can validate the algorithms using this data. At the end of the first reporting period, several iterations of filling out and answering inquiries about process documents have been carried out. It is foreseen that access to Scottish data will be in place in the second half of 2024, with plans for technical and clinical AICE team members to travel to Scotland to validate the AIAs locally, as it will not be possible to merge the Scottish data into a Danish or international research database.
Along with the practical tasks of getting access to data and completing lab tests and internal validation of the AICE algorithms, we have started the process of setting up of a clinical support system for how an AI-supported CCE service could be put into clinical practice. Analyses and workshops have led to identification of a "service data flow" diagram based on the current clinical practice and IT infrastructure at Odense University Hospital. In reporting period two and three these workflows will be further developed and fleshed out, working towards a final, recommended technical diagram for implementation into clinical practice, that takes into account the clinical reality and the use of AI.
A review study has been conducted to prepare the work on the diagnostic pathway that will be a main focus in reporting period two. As part of this work, patient and end-user representatives will be invited to contribute.
Preparations for the work on a 'CCE patient support mobile application' for patients undergoing a CCE procedure has also been started in Scotland, with plans for close patient inclusion in the design, development and test phases of the process.
This work will be carried out in close collaboration with the work on developing clinical implementation guidelines. The vision is to develop the guidelines as a set of living guidelines integrated in the app, collecting in one place all the scientific knowledge on both clinical CCE procedures and on AI-supported CCE used in clinical practice.
The main scientific impact of the first reporting period has been in relation to the technical AI algorithm development. Three scientific papers have been submitted for peer-reviewed publications; 1 published, 1 rejected in target journal (but will form part of another paper later in the project) and 1 still under review.
The results presented in the published paper are of substantial scientific importance in the field of AI development for healthcare. The method developed called AID-U-Net is a novel semantic segmentation technique that outperforms the current state-of-the-art solutions like U-Net, U-Net++ and W-Net. The performance of AID-U-Net is especially superior to others in low-resolution images such as those retrieved from CCE. The architecture of AID-U-Net is inspired by U-Net, but the addition of sub-contracting layers have significantly improved its performance.
The AID-U-Net method is applied in several of the algorithms developed as part of the AICE project.
A number of clinical scientific papers have also been written in the early stages of AICE, but not claimed as part of AICE, as they were part of other ongoing studies. The results of this work and ongoing AICE activities is currently being planned for scientific publication along with novel research around the ethical aspects of the project, development of the clinical diagnostic pathway and living guidelines of clinical implementation of AI-supported CCE diagnostics.
The results presented in the published paper are of substantial scientific importance in the field of AI development for healthcare. The method developed called AID-U-Net is a novel semantic segmentation technique that outperforms the current state-of-the-art solutions like U-Net, U-Net++ and W-Net. The performance of AID-U-Net is especially superior to others in low-resolution images such as those retrieved from CCE. The architecture of AID-U-Net is inspired by U-Net, but the addition of sub-contracting layers have significantly improved its performance.
The AID-U-Net method is applied in several of the algorithms developed as part of the AICE project.
A number of clinical scientific papers have also been written in the early stages of AICE, but not claimed as part of AICE, as they were part of other ongoing studies. The results of this work and ongoing AICE activities is currently being planned for scientific publication along with novel research around the ethical aspects of the project, development of the clinical diagnostic pathway and living guidelines of clinical implementation of AI-supported CCE diagnostics.