Revolutionizing Medical Imaging (ReImagine) through Ubiquitous, Low-Dose, Automated Computed Tomography Diagnostic Systems
Lead Research Organisation:
University of Cambridge
Department Name: Chemical Engineering and Biotechnology
Abstract
Imagine a world in which every individual can be routinely and extensively health monitored, in a time-efficient and safe manner, without having to visit an oversubscribed, centralised medical centre with limited access and appointment flexibility. Imagine a new clinical paradigm where early diagnosis becomes the standard, even in remote areas, within low-income demographics and for international travel, due to ubiquitous, modular, high-resolution X-ray imaging systems with automated diagnosis and live reporting; where frequent imaging contributes to a large diagnostic portfolio of individuals over time (whilst maintaining privacy) and advanced artificial-intelligence (AI)-based algorithms use these anonymous data sets acquired across the population to identify extremely early stages of disease - transforming preventative medicine as we know it. This is the 2050 that ReImagine will enable.
We will revolutionise the use of X-rays for medical imaging through low-dose, high-resolution and inexpensive computed tomography (CT) scanners, where highly innovative hardware and software components will be developed side-by-side to enable automated all-in-one pre-symptomatic diagnosis. Our vision will be enabled by developing highly sensitive X-ray detectors using scalable halide perovskite (PVK) semiconductors - materials currently making impact as disruptive photovoltaic (PV) technologies - for phase contrast X-ray imaging, in conjunction with AI-driven algorithms for image reconstruction, lesion detection and segmentation. This will realise quicker and more efficient healthcare delivery and prevent disease spread through extremely early detection of disease (e.g., those otherwise responsible for future pandemics) and for routine follow-up of oncology patients (e.g. early detection of cancer recurrence).
To realise this extremely challenging vision - combining breakthroughs in hardware, software and end-user application - we have uniquely assembled a world-leading, cross-cutting team from the Universities of Cambridge, Loughborough and Leicester, together with academic partners at the University of Leiden and industry partners in GE Healthcare, Scintacor, Cheyney and Immaterials Labs, bringing combined expertise spanning materials synthesis and scaling, characterisation and modelling, device assembly, detector physics, mathematics, CT systems development, and clinical radiology. The hardware will be interweaved with the software and algorithm development, with both guided by clinical insight, industry and case studies to ensure fit for end users.
We will revolutionise the use of X-rays for medical imaging through low-dose, high-resolution and inexpensive computed tomography (CT) scanners, where highly innovative hardware and software components will be developed side-by-side to enable automated all-in-one pre-symptomatic diagnosis. Our vision will be enabled by developing highly sensitive X-ray detectors using scalable halide perovskite (PVK) semiconductors - materials currently making impact as disruptive photovoltaic (PV) technologies - for phase contrast X-ray imaging, in conjunction with AI-driven algorithms for image reconstruction, lesion detection and segmentation. This will realise quicker and more efficient healthcare delivery and prevent disease spread through extremely early detection of disease (e.g., those otherwise responsible for future pandemics) and for routine follow-up of oncology patients (e.g. early detection of cancer recurrence).
To realise this extremely challenging vision - combining breakthroughs in hardware, software and end-user application - we have uniquely assembled a world-leading, cross-cutting team from the Universities of Cambridge, Loughborough and Leicester, together with academic partners at the University of Leiden and industry partners in GE Healthcare, Scintacor, Cheyney and Immaterials Labs, bringing combined expertise spanning materials synthesis and scaling, characterisation and modelling, device assembly, detector physics, mathematics, CT systems development, and clinical radiology. The hardware will be interweaved with the software and algorithm development, with both guided by clinical insight, industry and case studies to ensure fit for end users.
Organisations
- University of Cambridge (Lead Research Organisation)
- Rutherford Appleton Laboratory (Collaboration)
- Immaterial Labs (United Kingdom) (Project Partner)
- Science and Technology Facilities Council (Project Partner)
- Centrum Wiskunde & Informatica (Project Partner)
- Scintacor Ltd (Project Partner)
- General Electric (United States) (Project Partner)
- Cheyney (United Kingdom) (Project Partner)
Publications
Moseley ODI
(2021)
Halide perovskites scintillators: unique promise and current limitations.
in Journal of materials chemistry. C
Ferrer Orri J
(2022)
Unveiling the Interaction Mechanisms of Electron and X-ray Radiation with Halide Perovskite Semiconductors using Scanning Nanoprobe Diffraction.
in Advanced materials (Deerfield Beach, Fla.)
Landman M S
(2022)
On Krylov Methods for Large Scale CBCT Reconstruction
in Cornell University arXiv
Tumen-Ulzii G
(2022)
Defect Passivation by Pyridine-Carbazole Molecules for Efficient and Stable Perovskite Solar Cells
in ACS Applied Energy Materials
Jagt RA
(2023)
Layered BiOI single crystals capable of detecting low dose rates of X-rays.
in Nature communications
Dittmer S
(2023)
Navigating the challenges in creating complex data systems: a development philosophy
in Nature Machine Intelligence
Liu Y
(2023)
The Electronic Disorder Landscape of Mixed Halide Perovskites.
in ACS energy letters
| Description | We synthesised high-quality perovskite single crystals that meet the requirements for effective X-ray detection. We fabricated an X-ray detector with a record low dark current density of 10 pA cm-2 with single-photon-sensitivity which is critical to assemble into multi-pixel arrays beyond this project. This shows enormous promise for low dose X-ray detection to transform medical imaging. We assess against the objectives as follows: 1. Fabricate PVK composite materials with bulk resistivity of 1011 O cm, thicknesses of >100 µm Result: X-ray detectors were successfully constructed with 1000-µm-thick perovskite single crystals (objective: 100 µm; complete). Resistivity of the crystal is >109 O.cm (objective complete, where the value achieved is already suitable for combination with electronics). 2. Demonstrate deposition of pin-hole free, large-area (5 x 5 cm2) homogeneous films with thicknesses of >100 um using slot-die coating ready for scaling into larger devices. Result: We synthesised large and uniform (pin hole-free) perovskite single crystals with dimensions of 2x2 cm2 with thickness of 2000 um (objective: 5 x5 cm2; partially complete). This dimension fits with the commercially available chips (2 x 2 cm2) for the multi-pixel detection. 3. Fabricate a single pixel PVK direct detector (2.5 x 2.5 cm2) with dark current < 100 pA cm-2 (drift of < 5 % at 20 V bias) and lowest detectable X-ray dose rate of 15 nGyair s-1 at 100kVp with < 5 % increase in lowest detectable dose rate after X-ray exposure of 1120 Gyair. Result: We realised single-pixel perovskite detectors with extremely low dark current of 10 pA cm-2, 10x lower than our objective (<100 pA cm-2; complete). The detector shows a lowest detectable dose value of 167 nGyair s-1 at X-ray energy of 100 kVp X-ray energy (15 nGyair s-1 at 100 kVp; partially complete). The detector shows good stability under 100 kVp Xray exposure, with the X-ray induced current decreasing only 4.5% after dose of 273 Gyair (<5 % decrease after 1,120 Gyair, partially complete). 4. Construct a novel deep invertible reconstruction network trained on simulated and real data. Result: State of the art deep models were studied under varying low dose CT settings of this project, showing published methods have significant difference in performance (or lack of thereof) under different noise levels. Analysis for publication is being carried out (partially complete). Several ideas on more robust versions of existing methods are being currently developed (partially complete). 5. Perform task-adapted learned CT reconstruction for the segmentation of kidney lesions. Result: Research on task adapted inverse problems (for lung lesions, due to availability of data) has been performed. Finding reinforce the idea that task adapted metrics are required for model training and evaluation, as the models producing best image metrics do not produce images with highest clinical value (by e.g. deleting tumours in favour of smoothness). However, training for non-differentiable clinical parameters is not trivial. Further analysis is ongoing (partially complete) In essence, our research not only advanced the understanding of perovskite-based X-ray detectors but also provided a practical pathway towards developing highly sensitive and low-noise detectors capable of meeting the demanding requirements of modern X-ray imaging and detection technologies. Together, it also further the research of the gaps between theory and application of deep learning models for CT reconstruction. |
| Exploitation Route | Based on the results obtained in this phase, namely the achievement of low dark current and high sensitivity in perovskite single crystal-based X-ray detectors, there emerges a promising avenue for the development of multi-pixel single-photon counting X-ray detectors tailored for CT applications. The results will be useful for emerging startups and established medical imaging companies in this space, as well as academics and radiologists through further funded projects. The results from the machine learning project led to the current development of a toolbox of data-driven models for CT, focusing on reproducibility and real CT noise simulation. The current state of the art on these models is scattered and not easily comparable, and the toolbox attempts to breach this gap. It is now being developed further by our team. |
| Sectors | Digital/Communication/Information Technologies (including Software) Electronics Healthcare |
| Description | We have established a spin-out company to commercialise the results of the project. This is at early stage. |
| First Year Of Impact | 2022 |
| Sector | Electronics,Healthcare |
| Impact Types | Economic |
| Description | EPSRC Impact Acceleration Account Impact Grants 2022 |
| Amount | £61,718 (GBP) |
| Funding ID | RG90413 |
| Organisation | University of Cambridge |
| Sector | Academic/University |
| Country | United Kingdom |
| Start | 03/2022 |
| End | 06/2022 |
| Description | Ernest Oppenheimer Early Career Fellowship |
| Amount | £150,000 (GBP) |
| Organisation | University of Cambridge |
| Sector | Academic/University |
| Country | United Kingdom |
| Start | 09/2021 |
| End | 10/2022 |
| Description | Perovskite X-ray Photon Counting Detector |
| Amount | £200,512 (GBP) |
| Organisation | United Kingdom Research and Innovation |
| Sector | Public |
| Country | United Kingdom |
| Start | 03/2023 |
| End | 04/2025 |
| Title | Research data supporting "Unveiling the interaction mechanisms of electron and X-ray radiation with halide perovskite semiconductors using scanning nano-probe diffraction" |
| Description | The original data files for SED are found in the `data` folder. Each diffraction file can be opened using Python with some additional open-source packages (mainly pyXem). The nXRD and sXRD files are found in the data folder. They can be opened using Python (using hyperSpy). In the `models` folder, documents for modelling and simulations are stored. In the `casino_sim` there are the CASINO simulation for the electron beam interaction volume. In the `cross_sections` there are the calculations for the scattering cross sections. In the `crystal_structures` there are the collections of `.cif` files with the crystal structures of interest and the diffraction simulations that can be opened using the CrystalMaker software. In the `space_radiation` folder there are the SPENVIS simulation files. In the `notebooks` folder, all the Jupyter Notebook files are stored for the data analysis and data plotting of all the data files. In order to reproduce the results, please use the `requirements.txt` file to create an Anaconda virtual environment with all the necessary Python packages. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2022 |
| Provided To Others? | Yes |
| URL | https://www.repository.cam.ac.uk/handle/1810/335115 |
| Description | STFC-UKRI Project Partner |
| Organisation | Rutherford Appleton Laboratory |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | During this collaboration, we have successfully fabricated thick perovskite films devices, and developed a single-pixel detector. We have initially assessed their performances and single crystal-based devices have proven a better performance with very low dark currents of <0.001nA/cm2 and a reasonably good detection limit of 167 nGy(air)/s at 1000mV. Now, we aim for developing chips based on these devices and test their performance in collaboration with STFC. |
| Collaborator Contribution | STFC is contributing to the optimization of devices and their architecture. This is an iterative process in which we are also testing the performance of materials in its facilities. The complete iterative process implies synthesizing samples, producing devices, testing their performances both in our in-house equipments and in the STFC facilities. In addition, STFC is contributing to its know-how to produce photon counting ASICs chips, providing us with some, helping in the integration of the device and evaluating its performance. |
| Impact | We have successfully fabricated thick perovskite films based devices and measure their performance. We have also developed a backup approach for high quality perovskites single crystals, and integrate them into devices with better performances. We are currently in the process of optimising chips based on these approaches and we aim for producing a multi-pixel single photon counting detector. |
| Start Year | 2022 |
| Description | ReImagine Workshop |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Other audiences |
| Results and Impact | All different stakeholders from this project (both from academia and industry) participated on this workshop to disseminate objectives of our project, current status of its development and results, and making plans for the future of this project and to extend our collaboration. Thanks to this workshop, we attract the interest of the Detector Development Group (from Rutherford Appleton Laboratory, Science and Technology Facilities Council - STCF -), which eventually has joined as Project Partner and participates actively in research activities. |
| Year(s) Of Engagement Activity | 2022 |