Onwelo

Grant agreement number: FEMA.01.01-IP.01-02ED/24-00, implemented under Action 1.1 "Research, Development and Innovation of Enterprises", project type: "Modular Projects" of the European Funds for Mazovia 2021–2027, Priority I "European Funds for a More Competitive and Intelligent Mazovia", project co-financed by the European Regional Development Fund.

Project title: "Application of Advanced Information Technologies and Artificial Intelligence in Multiparametric Diagnostics of Prostate Cancer and Assessment of Multiple Sclerosis."

The research and development work planned within the project includes industrial research and experimental development work (in accordance with Article 2, point 85 of Commission Regulation (EU) No 651/2014 of 17 June 2014 declaring certain categories of aid compatible with the internal market in application of Articles 107 and 108 of the Treaty), and also leads to the implementation of their results into the beneficiary's business activity, contingent on a positive outcome of the R&D work.

Project duration: 18 months. Estimated duration for IR (Industrial Research): 12 months. The scope of planned work within IR is as follows:

Task 1: Development of a preliminary analysis and architecture for extending the main system with new plugin chains, along with MRI data analysis for the prostate:

1.1 Analysis of prostate MRI data with segmentations
1.2 Development of a preliminary analysis and architecture for extending the main system and creating plugin modules and chains
1.3 Development of a preliminary general specification of input/output interfaces for connecting plugins to the main system

The task will be based on analysis of data obtained from both publicly available datasets (e.g. Prostate-MRI-US-Biopsy, SPIE-AAPM-NCI PROSTATEx) and anonymised studies from medical facilities.

Task 2: Creation of a prototype algorithm for prostate segmentation into zones:

2.1 Development of the prostate zone segmentation algorithm
2.2 Expansion of the analysis and architecture for extending the main system and creating plugin modules and chains for the prostate
2.3 Development of input/output interface specifications for the prostate plugin
2.4 Development of automated tests and manual test scenarios for the prostate segmentation plugin, with initial test runs

Within this task, a prototype of deep learning-based algorithms for segmenting the prostate in MRI images and dividing it into zones will be developed. Adaptation of existing solutions will be possible at this stage; if satisfactory results are not achieved, proprietary tools will be developed. Deep learning methods are planned (e.g. UNet, VNet, ENet, FCN). This tool is essential for the project goals, as it will enable automatic prostate measurement and localisation of potential tumour lesions.

Task 3: Creation of a prototype algorithm for brain segmentation into zones:

3.1 Analysis of brain MRI data with segmentations
3.2 Development of the brain zone segmentation algorithm
3.3 Expansion of the analysis and architecture for extending the main system and creating plugin modules and chains for the brain
3.4 Development of input/output interface specifications for the brain segmentation plugin
3.5 Development and implementation of an adapter for the prostate segmentation plugin
3.6 Preliminary analysis and development of the main system data model expansion with new attributes returned by new plugins
3.7 Analysis and development of system extension points (e.g. dictionaries) for integrating new plugin functionality into the main system
3.8 Development of automated tests and manual test scenarios for the brain segmentation plugin, with initial test runs

At this stage, a tool for segmenting anatomical brain structures in MRI images is planned. Publicly available models (e.g. FastSurfer, MONAI Wholebrainseg) are planned to be adapted, verified on collected data, and potentially improved. The implementation of the anatomical segmentation algorithm will enable measurement of the volumes of different brain regions and analysis of the location of demyelinating plaques. Algorithms for detecting demyelinating plaques in the brain will then be developed, using deep learning methods such as UNet, UNet++, DeepLab, or transformer models, with detection based on FLAIR and T2-weighted MRI sequences, though additional sequences (e.g. T1-weighted, DIR, DWI) may be needed during algorithm development. Since most available MRI datasets of MS patients are released under licences that do not permit commercial use, the Beneficiary plans to work with subcontractors at this stage, including a medical unit specialising in MS and a research unit that will prepare precise outlines of demyelinating plaques. Subcontractors will be responsible for providing a database of anonymised studies for a minimum of 500 patients diagnosed with multiple sclerosis (500–1,500 studies, depending on availability and the progress of algorithm training), with precise demyelinating plaque outlines prepared for each patient.

Task 4: Creation of a prototype algorithm for detecting and assessing areas suspected of prostate cancer:

4.1 Development of the algorithm for detecting and assessing areas suspected of prostate cancer
4.2 Development and implementation of an adapter for the brain segmentation plugin
4.3 Development of input/output interface specifications for the prostate cancer detection and assessment plugin
4.4 Development and implementation of an adapter for the prostate cancer detection and assessment plugin
4.5 Development and implementation of a prototype adapter call chain for prostate and brain plugin sequences
4.6 Development and partial preliminary implementation of main system extensions for integrating newly developed plugin functionality
4.7 Development of automated tests and manual test scenarios for the prostate plugin at this stage, with initial test runs

The aim of this task is to develop algorithms for detecting tumour lesions in the prostate. Deep learning methods are planned, such as UNet, ResNet, Attention U-Net, or transformer models, with detection based on combined information from multiple MRI sequences (e.g. T2-weighted, T1-weighted, DWI).

Task 5: Creation of a prototype algorithm for detecting and measuring changes associated with multiple sclerosis:

5.1 Development of the algorithm for detecting and measuring changes associated with multiple sclerosis
5.2 Development of input/output interface specifications for the MS detection and measurement plugin
5.3 Development and implementation of an adapter for the MS detection and measurement plugin
5.4 Expansion and finalisation of the adapter call chain for prostate and brain plugin sequences
5.5 Continued development and preliminary implementation of main system extensions for newly developed plugin functionality
5.6 Preliminary implementation of the main system data model expansion with new attributes returned by new plugins
5.7 Development of automated tests and manual test scenarios for MS functionality, with initial test runs

The aim of this task is to develop algorithms for detecting demyelinating plaques in the brain. Deep learning methods are planned, such as UNet, UNet++, DeepLab, or transformer models, with detection based on FLAIR and T2-weighted MRI sequences, though additional sequences (e.g. T1-weighted, DIR, DWI) may be needed during algorithm development. Once trained, the model will enable assessment of the number and location of plaques.

Estimated duration for ED (Experimental Development): 5 months. The scope of planned work within ED is as follows:

Task 6: Integration of plugins with the central system message exchange mechanism:

6.1 Expansion and finalisation of main system extensions for newly developed plugin functionality
6.2 Implementation and finalisation of the main system data model expansion with new attributes returned by new plugins
6.3 Implementation of business process extensions to invoke new plugins
6.4 Development of automated tests and manual test scenarios for new end-to-end process functionality, with test runs
6.5 End-to-end tests and test reports for newly developed functionality

Within this task, the developed algorithms will be integrated with the Raygenic DICOM viewer. At this stage, a plugin compatible with the viewer and supporting the agreed input/output contracts will need to be prepared.

Task 7: Final technology verification:

7.1 Mandatory testing of the software/AI algorithm in an isolated environment by physicians (subcontractor task)
7.2 End-to-end tests and test reports in an isolated environment for newly developed functionality
7.3 Finalisation of the analysis and architecture documentation for the main system expansion with new plugins
7.4 Support of the testing process through development, corrective, analytical and architectural activities arising from identified deficiencies
7.5 Verification and completion of documentation for compliance with Polish law and supervision of medical validation (subcontractor task)

Within this task, the developed solution will be tested on test datasets containing data not used at any stage of training, in an environment close to real-world conditions. The validation stage is critical: before algorithms are approved for use by physicians on real data, it must be confirmed that they operate correctly and meet the required thresholds of accuracy, sensitivity and specificity. A subcontractor will be engaged to verify and complete the documentation for compliance with Polish law and to supervise medical validation. Hospital environments are also planned to be involved in software testing, with results interpreted by researchers from a medical university, covering both prostate cancer and MS.

Estimated duration for PW (Pre-implementation Work): 1 month. The scope of planned work within PW is as follows:

Task 8: Pre-implementation work:

8.1 Market research and preparation of sales documentation, including contract templates and regulations concerning the implementation of systems in hospitals and personal data security (GDPR)

A subcontractor with experience in market research and public opinion research will be engaged.

The project addresses specific challenges related to current limitations in medical imaging diagnostics, particularly in the context of prostate cancer and multiple sclerosis (MS). There is insufficient availability of specialists, leading to an overburdened healthcare system and delays in diagnosis and treatment. In response to these needs, the project focuses on developing advanced algorithms for automatic MRI image analysis to accelerate the diagnostic process and increase the objectivity of radiological assessment.

For prostate cancer, the project involves creating a tool for automatic prostate segmentation and detection of areas suspected of tumour lesions, with automatic assessment of the clinical significance of potential tumour changes in accordance with PI-RADS guidelines. For multiple sclerosis, the project involves developing a tool for automatic brain segmentation, detection of demyelinating plaques, and monitoring of disease progression.

The project assumes that following implementation of the planned R&D work, the beneficiary will become competitive in the domestic market by offering innovative tools supporting the diagnosis of prostate cancer and multiple sclerosis. Market competitiveness will stem from several factors:

Technological innovation: the developed tools will use advanced machine learning algorithms and image processing techniques to provide fast, precise, and automatic MRI analysis in line with medical guidelines.

Effectiveness and efficiency: faster diagnosis and treatment will allow medical facilities to increase productivity and improve service quality. Standardisation and objectification of diagnostics: automatic MRI analysis will contribute to standardising the diagnostic process and increasing the objectivity of radiological assessment.

Cost savings: faster diagnosis and treatment can reduce costs associated with lengthy diagnostic processes and ineffective therapies.

The goal of the project by Onwelo Spółka z ograniczoną odpowiedzialnością (limited liability company) is to develop a tool supporting diagnostics and decision-making in the treatment of prostate cancer patients — in the area of detection and assessment of the malignancy of prostate tumours — and in the treatment of multiple sclerosis patients — in the area of detection and assessment of demyelinating lesions in the brain, assessment of brain atrophy, and monitoring of disease progression using magnetic resonance imaging (MRI). Within the R&D work, the Beneficiary intends to use machine learning methods, in particular deep learning, including convolutional neural networks, which demonstrate exceptional effectiveness in analysing complex patterns in medical images, as well as classical image processing methods such as filtering, threshold-based segmentation, and morphological analysis.

Target groups: hospital urology, oncological urology and neurology wards; specialist urology and neurology outpatient clinics; diagnostic imaging departments; pathomorphology departments; general practitioners; private practices of specialist urologists and neurologists; and private radiology clinics. Through automatic analysis of imaging studies to identify, analyse and describe potential lesions, physicians can assess more study results during their work, which translates into an optimised treatment process and increased availability and efficiency of healthcare.

Project value: PLN 4,164,469.42 net (PLN 4,461,068.76 gross).

European Funds contribution: PLN 1,968,550.42 (47%).

Project implementation period: 01/10/2024 – 31/03/2026.