Medical imaging markers of cancer initiation, progression and therapeutic response in the breast based on tissue microstructure

Lead Research Organisation: University College London
Department Name: Medical Physics and Biomedical Eng

Abstract

Studies of cancer DNA show that breast cancer comprises many different diseases. Some spread quickly and are very dangerous, while others are slow growing and are not life-threatening. Some respond well to chemotherapy or radiotherapy, while others are resistant to treatment. It is important to be able to select the best treatment for a particular patient's cancer, so-called "patient stratification". In this way appropriate treatment can be given whilst potential side-effects from an ineffective treatment are avoided.

Recent discoveries have implicated the stroma, the connective tissue of the breast, in both the aggressiveness of breast cancer and its response to therapy. When small amounts of tissue are removed from a potential breast cancer by needle biopsy, or when a breast cancer is removed by surgery, subtle differences are seen in microscopic imaging (histology) of the stromal tissue around the tumour. These occur at the scale of a few microns and are far below the spatial resolution of X-ray mammography, magnetic resonance imaging (MRI) and ultrasound that radiologists routinely use to detect and assess breast cancer.

A recent innovation, ultrasound shear wave elastography (USSWE), has enabled radiologists in Dundee to show differences in stromal tissue of patients with invasive cancer. USSWE measures the very small displacements of tissue that occur during conventional ultrasound examinations. The MR imaging is specifically designed to (a) assess the microscopic blood supply of the breast tissue, and any cancer present, using a method called dynamic contrast enhancement, DCE-MRI, and (b) assess the size and shape of microstructure by measuring the diffusion of protons, DWI-MRI. Accurate comparison of these with histology of the removed tissue enables the computer to learn the differences in the images between normal tissue and cancer and between normal stroma and stroma associated with cancer. In a further step scientists at the UCL Centre for Medical Image Computing have invented new methods to measure some of these microscopic changes. They generate computer models of these structures and use the subtle changes predicted by these models to adjust the imaging techniques to maximise the differences between normal tissue and the different types of cancer and stromal tissue changes that we see using histology.

We propose a project in which we will develop ways to combine X-ray DBT, DCE-MRI, DWI-MRI and USSWE to best discriminate between normal stroma and stroma near to cancers, and between the different types of cancer: high risk, low risk, and those that do and do not respond to therapy. We will first ask 3 women who have recently been diagnosed with breast cancer to spend about an hour in a MR scanner to allow us to collect a wide range of DW-MRI images so we can optimise MRI acquisition. We will then ask a further 30 women recently diagnosed with cancer for permission to use their radiological images and the histology images of their tissue samples from biopsy and surgery. We will use these images to train our computer methods to best discriminate between the different types of stroma and cancer, and between the different components of the cancer itself. Ten of the patients selected will be undergoing a course of neo-adjuvant therapy (a type of chemotherapy that is given prior to surgery to shrink the tumour).

If successful this project will discover new ways to identify those patients who will respond best to a particular treatment, those patients who do not have life threatening disease and so do not need aggressive therapy, and those patients with types of cancer that require more aggressive treatment. At the end of this project we will translate these findings to a clinical trial.

Planned Impact

Breast cancer represents a spectrum of disease, from lesions which are so indolent that they would never present clinically or threaten life, to cancers which - despite aggressive multimodal treatment including surgery, radiotherapy and chemotherapy - will cause death in a short period of time. Although standard prognostic features can sometimes separate breast cancer patients into low and high risk groups, actual prognostication and prediction of treatment responsiveness for many individual patients remains poor. As a result, overtreatment is common, since the consequences of overtreatment are considered to be less adverse than those of under-treatment.

The outcomes of this work, if successful at subsequent clinical trial, will help breast cancer patients and health care professionals' desire for personalised and stratified healthcare to become a reality, with the long term aim of a more accurate disease prognostication and treatment response prediction. This should lead to less overtreatment - with expensive and toxic therapies - of women with low risk disease, and to less risk of under-treatment of women with high risk disease. The further stratification of women at high risk of breast cancer death into those who will be either responsive or resistant to particular therapies will benefit:
a) breast cancer patients who can balance the harms and benefits of treatments offered
b) healthcare providers who will be better able to manage scant resources
c) treatment developers, manufacturers and providers, who may be able to target treatments more effectively to specific subgroups when such treatments may not be cost effective for larger and more heterogeneous patient cohorts.

These benefits will be particularly evident for women with oestrogen receptor negative (ER-) breast cancer, where useful predictive markers are currently lacking. Such cancers tend to be aggressive and occur in younger women, so better management of such disease would have a profound positive impact on society.

Eventually, the integration of multimodal imaging prognostic and predictive information into prognostic and predictive indices could be routinely exploited through the derivation and inclusion of such indices into multimodal imaging workstations. The integration of such imaging parameters with histological, immunohistochemical and genetic information could be developed further into a commercial online prognostic and predictive tool (such as a next generation successor to the currently widely used "Adjuvant Online!" website).

The project partners all have strong links with industry and commercialisation via licensing, consultancy or spin-out company formation is the only really effective way to disseminate the results of new methodological research widely to healthcare providers. We will protect new ideas by patent as appropriate through the commercialisation arms of the two institutions.
We will also disseminate the results of our work through the extensive public engagement and patient awareness initiatives at our disposal and documented in our Pathways to Impact document.

Publications

10 25 50
 
Description We started the project with the hypothesis that there might be changes in the peritumoral stroma, that might be indicators of cancer risk, and that these indicators might be detectable using specialised adaptations of common imaging modalities, in particular MRI and ultrasound. This is important as this might help breast cancer specialists to assess, non-invasively, whether a detected cancer is high risk or low risk. This could in turn address the significant problem of over treatment in common cancers such as those of the breast or prostate. The proposal concerned the diagnosis and risk assessment in breast cancer but the same concepts are also being explored in prostate cancer.

The initial work was started with resource from the EPSRC Programme "Intelligent Imaging: Motion, Form and Function across Scale" during which the model based diffusion MRI method, VERDICT, was applied to prostate cancer. VERDICT provides a parsimonious model based on simple geometric descriptors of tissue microstructure such as spheres, ellipsoids and rods to explain the MR diffusion signal (Panagiotaki et al 2014). The dimensions of these elements described the tissue microstructure and can be interpreted in terms of cell size, the volume of extracellular space, cell packing and a vascular component. The VERDICT work follows the imaging biomarker roadmap for cancer studies (O'Conner et al 2016) that the PI contributed to.

Our project comprised 6 components that were presented as 4 workpackages in the original proposal.

1. Collection of data

Our collaborators in Dundee obtained multi-parametric MRI, including diffusion MRI with multiple b-values on 19 patients. Shear wave elastography ultrasound images were obtained on 1137 tumours from 1112 women. A specific MRI protocol was devised and tested to assess tissue microstructure using VERDICT and a specific ultrasound elastography acquisitions were obtained to assess the mechanical properties of the peritumoral stroma.

2. Methods for aligning histology with medical imagery

This followed from our successful work aligning X-ray mammogram with multiparametric MRI of the breast (Mertzanidou et al 2014) that was finished off at the beginning of this project. Alignment with histology proved to be very challenging in breast cancer but with the help of pathologists and physicists in Nijmegen we developed a sample preparation, imaging and alignment protocol as part of the EU VP7 "PRISM" project. With data on 5 patients who underwent mastectomy we devised a method for producing a coherent 3D volume of histology data that can be related back to post-surgery 3D imaging (MR or CT). This work has been presented at two IWDM meetings (Mertzanidou et al 2014, 2016) and is submitted for publication. Initial work has also been completed using the same method for lumpectomy samples. MRI has also been successfully aligned to histology of 7 breast cancer samples obtained from the King's Health Partners Cancer Biobank.

This is an important area of work and there is still much to do, so we have significantly expanded effort in this area. Most advanced is our work on prostate cancer where Laura Panagiotaki (EPSRC Research Fellow) has developed a novel method to prepare and slice whole prostatectomy samples to aid correlation between clinical and histological images. We are now developing a method to align histology data with clinical imaging data based on statistical measures of alignment (normalised mutual information) and are looking to extend this with a machine learning approach.

We are also about to start a project on the liver assessing the effect of intra-vascular embolization and therapeutic efficacy after Trans-Arterial Catheter Embolization (TACE) followed by liver resection.

3. Diffusion MRI

a. Clinical data

Of the 19 patients whose data was collected there were 23 lesions. Three had just had chemotherapy and for two of these we had histology. Fifteen had surgery and all these had histology. On examination of the DCE MRI 10 patients had lesions of sufficient size (>1cm) for further analysis. These were the 3 with chemotherapy and 7 of the patients with surgery, yielding 9 with imaging and histology for our study. A detailed analysis of these data (Bailey et al ISMRM 2015) showed that VERDICT better explained the MR diffusion signal than ADC or the IVIM model (Sigmund et al 2011) in 61% of the tumour rim compared and 36% at the centre, although many tumours demonstrated significant heterogeneity. This initial study demonstrates the potential of the method but further studies are required with an MR system that provides the highest b values to fully determine whether these descriptors are in any way predictive of cancer risk.

b. Analysis of ex-vivo breast cancer tissue sample data
The initial clinical study prompted us to undertake a more controlled experiment using fixed tissue from 7 tumours obtained after surgery on 6 patients and curated as part of the King's Health Partners Cancer Biobank with the assistance of Sarah Pinder. These specimens were scanned on a 9.4T pre-clinical MR system at the Centre for Advanced Biomedical Imaging at UCL and histology slides prepared of representative regions at Guy's Hospital. Models with restricted diffusion components and anisotropy fitted the data best in the most cancerous regions and there were no regions where conventional ADC or DT models fitted the data well. Maps of parameters from these more complex models suggest that both overall cell volume fraction and individual cell size can contribute to the diffusion signal, explaining the poorer specificity ADC has for explaining tissue microstructure. After careful registration of histology and MRI the orientation of diffusion anisotropy corresponded to stromal orientation patterns on histology. A paper on this work has been submitted for publication.

4. Shear Wave Ultrasound Elastography (SWE) of the Breast

An early motivation of this project was the observation by Prof Andy Evans that the peritumoral stroma appears stiffer than stroma far from the lesion in SWE. A large number of clinical scans were collected with 1137 tumours from 1112 women. The resulting study (Evans et al 2016) showed that most breast cancer histological types have similar SWE characteristics. The exception was tubular cancer which has significantly lower stiffness than other histologic types, accounted for largely by their small size.

5. Analysis of histological data

a. Analysis of peritumoral stroma in breast cancer

Sara Ferreira-Reis is near completion of a PhD on the topic of histological image analysis to support this project. There is evidence that the structure of peritumoral stroma is indicative of cancer risk. The hypothesis is that slower growing cancers result in immature stroma that has reacted to the growing cancer while the most aggressive cancers just show physical distortion. Sara has explored the effect of histology slice spacing on the integrity of 3D reconstructions (Reis et al SPIE 2015) and devised a method for automatically classifying and quantifying stromal maturity (Reis et al SPIE 2016, paper under review). Her preliminary analysis has shown a statistically significant relationship between her texture measures and cancer grade. This is an important discovery and links to the findings on the breast cancer tissue sample data described above, raising the exciting possibility that MR diffusion imaging may be able to pick up these signals in-vivo. Testing this will be the next step.

6. Tumour modelling

Core to the work described above is better understanding of the biophysics of tumour growth, vascularity and the effect on surrounding stroma. To this end we have embarked on an ambitious programme to develop an appropriate biophysical model of tumour growth. This has been led by Vasileios Vavourakis over the last two years, supported by a Marie Curie Fellowship. Pete Wijeratne et al have developed a multi-scale model of tumour growth with a particular emphasis on the tumour-stroma boundary and the simulation of collagen remodelling as a reaction to tumour growth (Wijeratne et al 2016). This initial piece of work has shown how tumour growth can lead to passive mechanical re-orientation of collagen fibres and that peritumoral fibre network anisotropy can lead to anisotropic tumour morphology. In other work by Vavourakis, submitted for publication, a multiscale in-silico model for mechano-sensitive tumour angiogenesis and growth has been developed. Related work in wound healing has been used to successfully predict the cosmetic effect of the breast after surgical lumpectomy to treat breast cancer (Vavourakis et al 2016). This work was primarily funded by the EU FP7 project "PICTURE" with contributions from this grant. This piece of work is likely to have significant clinical impact in the short to medium term.

Our team has written a review of the application of tissue biomechanics to breast image registration (Hipwell et al 2016).

Bibliography

For bibliography please see the publications reported on Researchfish
Note - several of the papers in preparation above have now been published.
Rachel Denholm, Bianca L De Stavola, John H Hipwell, Simon J Doran, Jeff MP Holly, Elizabeth Folkerd, Mitch Dowsett, Martin O Leach, David J Hawkes, Isabel dos-Santos-Silva "Circulating Growth and Sex Hormone Levels and Breast Tissue Composition in Young Nulliparous Women", Cancer Epidemiology and Prevention Biomarkers, 2018, 27, 1500-1508
Rachel Denholm, Bianca De Stavola, John H Hipwell, Simon J Doran, Marta C Busana, Martin O Leach, David J Hawkes, Isabel dos-Santos-Silva "Growth Trajectories, Breast Size, and Breast-Tissue Composition in a British Prebirth Cohort of Young Women", American J Epidemiology, 2017, 187, 1259-1268
Wijeratne PA, Hipwell JH, Hawkes DJ, Stylianopoulos T, Vavourakis V "Multiscale biphasic modelling of peritumoural collagen microstructure: The effect of tumour growth on permeability and fluid flow" PLoS one, 2017, 12, e0184511
Doran SJ, Hipwell JH, Denholm R, Eiben B, Busana M, Hawkes DJ, Leach MO, dos Santos Silva I, "Breast MRI segmentation for density estimation: Do different methods give the same results and how much do differences matter?", Medical Physics 2017, 44, 4573-4592
Mertzanidou T, Hipwell JH, Reis S, Hawkes DJ, Bejnordi BE, Dalmis M, Vreemann S, Platel B, Laak J, Karssemeijer N, Hermsen M, Bult P, Mann R, "3D volume reconstruction from serial breast specimen radiographs for mapping between histology and 3D whole specimen imaging", Medical Physics, 2017, 44, 935-948
Reis S, Gazinska P, Hipwell J, Mertzanidou T, Naidoo K, Williams N, Pinder S, Hawkes DJ, "Automated Classification of Breast Cancer Stroma Maturity from Histological Images", IEEE Transactions on Biomedical Engineering, 2017, 64, 2344-2352
Bailey C, Siow B, Panagiotaki E, Hipwell JH, Mertzanidou T, Owen J, Gazinska P, Pinder SE, Alexander DC, Hawkes DJ, "Microstructural models for diffusion MRI in breast cancer and surrounding stroma: an ex vivo study", NMR in Biomedicine, 2017, 30, e3679.
Vavourakis V, Wijeratne PA, Shipley R, Loizidou M, Stylianopoulos T, Hawkes DJ, "A Validated Multiscale In-Silico Model for Mechano-sensitive Tumour Angiogenesis and Growth", PLoS computing biology, 2017, 12, e0168317
O'Connor, J.P.B., Aboagye, E.O., Adams, J.E., Aerts, H.J.W.L., Barrington, S.F., Beer, A.J. et al. (Hawkes DJ one of 78 authors) "Imaging Biomarker Roadmap for Cancer Studies" Nat Rev Clin Oncol 14 (3), 169 doi:10.1038/nrclinonc.2016.162 (2017).

A patent that derived from some of the work done in this project has been filed by our collaborators, Philips, and has been granted in the UK, Germany, France, Russia, Canada, US, Japan. WO2017148702A1 DEVICE, IMAGING SYSTEM AND METHOD FOR CORRECTION OF A MEDICAL BREAST IMAGE.
Exploitation Route Some of our registration and simulation software has been made open-source with plans for more of it to go open source in due course.

As stated in the report above we are extending the concepts to other cancers, most notably prostate, with plans for assessing response to therapy in liver cancer.

Specific ideas following from our work has been adopted and patented by Philips.
Sectors Digital/Communication/Information Technologies (including Software),Healthcare

 
Description Our work has been reported to both Cancer Research UK and Prostate Cancer UK with related follow on funding from each. It is mentioned indirectly in the campaign to make multi-parametric MRI more widely available for cancer detection and staging, in particular in prostate cancer. Our work in modelling breast deformation has been taken up by Philips who have been granted a patent jointly with us: WO2017148702A1 DEVICE, IMAGING SYSTEM AND METHOD FOR CORRECTION OF A MEDICAL BREAST IMAGE. This is likely to contribute to applications in breast MR imaging.
First Year Of Impact 2019
Sector Healthcare,Manufacturing, including Industrial Biotechology
Impact Types Societal,Economic,Policy & public services

 
Description FP7 (PICTURE)
Amount € 735,000 (EUR)
Organisation European Commission 
Sector Public
Country European Union (EU)
Start 02/2013 
End 01/2016
 
Description FP7 (PRISM)
Amount € 738,000 (EUR)
Funding ID 316746 
Organisation European Commission 
Sector Public
Country European Union (EU)
Start 03/2013 
End 02/2016
 
Description Marie Curie Fellowship - towards the simulation of Breast...
Amount € 231,000 (EUR)
Funding ID 627025 
Organisation European Commission 
Sector Public
Country European Union (EU)
Start 02/2015 
End 01/2017
 
Description Philips
Amount £119,000 (GBP)
Organisation Koninklijke Philips Electronics N.V. 
Department Philips
Sector Private
Country Global
Start 09/2010 
End 08/2014
 
Title Report on MIMIC 
Description Report on research outcomes of project: We started the project with the hypothesis that there might be changes in the peritumoral stroma, that might be indicators of cancer risk, and that these indicators might be detectable using specialised adaptations of common imaging modalities, in particular MRI and ultrasound. This is important as this might help breast cancer specialists to assess, non-invasively, whether a detected cancer is high risk or low risk. This could in turn address the significant problem of over treatment in common cancers such as those of the breast or prostate. The proposal concerned the diagnosis and risk assessment in breast cancer but the same concepts are also being explored in prostate cancer. The initial work was started with resource from the EPSRC Programme "Intelligent Imaging: Motion, Form and Function across Scale" during which the model based diffusion MRI method, VERDICT, was applied to prostate cancer. VERDICT provides a parsimonious model based on simple geometric descriptors of tissue microstructure such as spheres, ellipsoids and rods to explain the MR diffusion signal. The dimensions of these elements described the tissue microstructure and can be interpreted in terms of cell size, the volume of extracellular space, cell packing and a vascular component. The VERDICT work follows the imaging biomarker roadmap for cancer studies (O'Conner et al 2016) that the PI contributed to. Our project comprised 6 components that were presented as 4 workpackages in the original proposal. 1. Collection of data Our collaborators in Dundee obtained multi-parametric MRI, including diffusion MRI with multiple b-values on 19 patients. Shear wave elastography ultrasound images were obtained on 1137 tumours from 1112 women. A specific MRI protocol was devised and tested to assess tissue microstructure using VERDICT and a specific ultrasound elastography acquisitions were obtained to assess the mechanical properties of the peritumoral stroma. 2. Methods for aligning histology with medical imagery This followed from our successful work aligning X-ray mammogram with multiparametric MRI of the breast (Mertzanidou et al 2014) that was finished off at the beginning of this project. Alignment with histology proved to be very challenging in breast cancer but with the help of pathologists and physicists in Nijmegen we developed a sample preparation, imaging and alignment protocol as part of the EU VP7 "PRISM" project. With data on 5 patients who underwent mastectomy we devised a method for producing a coherent 3D volume of histology data that can be related back to post-surgery 3D imaging (MR or CT). This work has been presented at two IWDM meetings (Mertzanidou et al 2014, 2016) and a journal paper has just been accepted (Mertzanidou et al 2017). Initial work has also been completed using the same method for lumpectomy samples. MRI has also been successfully aligned to histology of 7 breast cancer samples obtained from the King's Health Partners Cancer Biobank (see below). This is an important area of work and there is still much to do, so we have significantly expanded effort in this area. Most advanced is our work on prostate cancer where Laura Panagiotaki (EPSRC Research Fellow) has developed a novel method to prepare and slice whole prostatectomy samples to aid correlation between clinical and histological images. We are now developing a method to align histology data with clinical imaging data based on statistical measures of alignment (normalised mutual information) and are looking to extend this with a machine learning approach. We are also about to start a project on the liver assessing the effect of intra-vascular embolization and therapeutic efficacy after Trans-Arterial Catheter Embolization (TACE) followed by liver resection. 3. Diffusion MRI a. Clinical data Of the 19 patients whose data was collected there were 23 lesions. Three had just had chemotherapy and for two of these we had histology. Fifteen had surgery and all these had histology. On examination of the DCE MRI 10 patients had lesions of sufficient size (>1cm) for further analysis. These were the 3 with chemotherapy and 7 of the patients with surgery, yielding 9 with imaging and histology for our study. A detailed analysis of these data (Bailey et al ISMRM 2015) showed that VERDICT better explained the MR diffusion signal than ADC or the IVIM model (Sigmund et al 2011) in 61% of the tumour rim compared and 36% at the centre, although many tumours demonstrated significant heterogeneity. This initial study demonstrates the potential of the method but further studies are required with an MR system that provides the highest b values to fully determine whether these descriptors are in any way predictive of cancer risk. b. Analysis of ex-vivo breast cancer tissue sample data The initial clinical study prompted us to undertake a more controlled experiment using fixed tissue from 7 tumours obtained after surgery on 6 patients and curated as part of the King's Health Partners Cancer Biobank with the assistance of Sarah Pinder. These specimens were scanned on a 9.4T pre-clinical MR system at the Centre for Advanced Biomedical Imaging at UCL and histology slides prepared of representative regions at Guy's Hospital. Models with restricted diffusion components and anisotropy fitted the data best in the most cancerous regions and there were no regions where conventional ADC or DT models fitted the data well. Maps of parameters from these more complex models suggest that both overall cell volume fraction and individual cell size can contribute to the diffusion signal, explaining the poorer specificity ADC has for explaining tissue microstructure. After careful registration of histology and MRI the orientation of diffusion anisotropy corresponded to stromal orientation patterns on histology. A paper on this work has just been accepted (Bailey et al 2017) and is generating significant interest. 4. Shear Wave Ultrasound Elastography (SWE) of the Breast An early motivation of this project was the observation by Prof Andy Evans that the peritumoral stroma appears stiffer than stroma far from the lesion in SWE. A large number of clinical scans were collected with 1137 tumours from 1112 women. The resulting study (Evans et al 2016) showed that most breast cancer histological types have similar SWE characteristics. The exception was tubular cancer which has significantly lower stiffness than other histologic types, accounted for largely by their small size. 5. Analysis of histological data a. Analysis of peritumoral stroma in breast cancer Sara Ferreira-Reis is near completion of a PhD on the topic of histological image analysis to support this project. There is evidence that the structure of peritumoral stroma is indicative of cancer risk. The hypothesis is that slower growing cancers result in immature stroma that has reacted to the growing cancer while the most aggressive cancers just show physical distortion. Sara has explored the effect of histology slice spacing on the integrity of 3D reconstructions (Reis et al SPIE 2015) and devised a method for automatically classifying and quantifying stromal maturity (Reis et al SPIE 2016, Reis et al 2017). Her preliminary analysis has shown a statistically significant relationship between her texture measures and cancer grade. This is an important discovery and links to the findings on the breast cancer tissue sample data described above, raising the exciting possibility that MR diffusion imaging may be able to pick up these signals in-vivo. Testing this will be the next step. 6. Tumour modelling Core to the work described above is better understanding of the biophysics of tumour growth, vascularity and the effect on surrounding stroma. To this end we have embarked on an ambitious programme to develop an appropriate biophysical model of tumour growth. This has been led by Vasileios Vavourakis over the last two years, supported by a Marie Curie Fellowship. Peter Wijeratne et al have developed a multi-scale model of tumour growth with a particular emphasis on the tumour-stroma boundary and the simulation of collagen remodelling as a reaction to tumour growth (Wijeratne et al 2016). This initial piece of work has shown how tumour growth can lead to passive mechanical re-orientation of collagen fibres and that peritumoral fibre network anisotropy can lead to anisotropic tumour morphology. In other work by Vavourakis, submitted for publication, a multiscale in-silico model for mechano-sensitive tumour angiogenesis and growth has been developed. Related work in wound healing has been used to successfully predict the cosmetic effect of the breast after surgical lumpectomy to treat breast cancer (Vavourakis et al 2016). This work was primarily funded by the EU FP7 project "PICTURE" with contributions from this grant. This piece of work is likely to have significant clinical impact in the short to medium term. Our team has written a review of the application of tissue biomechanics to breast image registration (Hipwell et al 2016). 
Type Of Material Data analysis technique 
Year Produced 2016 
Provided To Others? Yes  
Impact See above. Scientific impacts are provided in the publications resulting from this work. Most notable impacts are: Methods for aligning histology data with medical imaging data (Mertzanidou et al 2017) Correlation of breast MR diffusion anistropy with peritumoral stromal patterns (Bailey et al 2017) Analysis of peritumoral stromal patterns and correlation with cancer risk (Reis et al 2017) Modelling peritumoral collagen re-modelling and re-orientation in high risk cancer (Wjeratne et al 2016) Preliminary work on modelling wound healing post lumpectomy (Vavourakis et al 2016) 
 
Description Berlin 
Organisation Charité - University of Medicine Berlin
Department Institute of Radiology
Country Germany 
Sector Hospitals 
PI Contribution Image registration technologies for alignment of X-ray mammograms with breast MRI images
Collaborator Contribution Carefull assessment of the results of our alignment software through a radiological observer study
Impact Alignment software and a clinical validation for establishing correspondence between X-ray mammograms and breast MR images (Martzanidou et al 2012 and 2014).
Start Year 2008
 
Description Dundee Radiology 
Organisation University of Dundee
Country United Kingdom 
Sector Academic/University 
PI Contribution As part of this EPSRC project we provided advice and assistance on diffusion MRI sequences for breast MR imaging; advice as how to prepare lumpectomy samples for comparision between post-op histology and pre-op MR imaging; input to analysis of ultrasound shear wave data including results from our numerical modelling work.
Collaborator Contribution Our Dundee partners, Professor Andy Evans and Dr Sarah Vinnicombe, provided multi-parametric MR images including multiple b-value diffusion imaging to our specification on 19 patients. They also provided access to shear wave ultrasound images on 1137 tumours from 1112 women. Both sides contributed in a major way to the development of the ideas and the study protocols of the project.
Impact A number of research presentations and journal papers have arisen from this work. Please see the reference list. The Impact has been to furthering the clinical and scientific goals of our project. There doesn't seem to be a box available for that below.
Start Year 2012
 
Description Frauhnhofer Mevis 
Organisation Fraunhofer Society
Department Fraunhofer MEVIS
Country Germany 
Sector Charity/Non Profit 
PI Contribution As partners in two EU projects (HAMAM and PRISM) we have supplied expertise in image registration, biomechanical modelling and MR diffusion imaging
Collaborator Contribution Mevis provided expertise in software developments, clinical translation, visualisation and image alignment.
Impact Various papers and conference publications (see publication list). Mevis produced several radiology systems for breast image integration, analysis and visualisation in which we had input.
Start Year 2008
 
Description Guys Pathology Collaboration 
Organisation King's College London
Department Division of Imaging Sciences and Biomedical Engineering
Country United Kingdom 
Sector Academic/University 
PI Contribution We have worked closely with Professor Sarah Pinder and her team in breast pathology. We have provided the output of our work which is an academic image analysis system for assessing peritumoral breast stroma maturity. We have worked closely to together to devise a novel image analysis system for undertaking this task (Reis et al 2017). We have assessed the VERDICT MR diffusion analysis on the 7 breast cancer samples supplied by KCL. Data has been correlated accurately with histology. Detailed analysis of the diffusion signal and it's anistropy has enabled significant advances in our understanding and interpretation of the MR diffusion signal in terms of tissue microstructure (Bailey et al 2017).
Collaborator Contribution Sarah Pinder has provided invaluable advice and expertise in pathological image interpretation. She and her team have undertaken a significant observer study to evaluate our methods. They have also supplied us with 7 breast cancer samples from the King's Health Partners Cancer Biobank.
Impact This is a multidisicplinary collaboration between breast pathology and medical image computing. Several papers have resulted from this work - see publication list. Again the impact is scientific and medical - not one of the boxes below.
Start Year 2012
 
Description Lisbon 
Organisation University of Lisbon
Country Portugal 
Sector Academic/University 
PI Contribution We have provided visualisation, image analysis and image registration expertise to the breast surgery group in Lisbon as part of teh PICTURE project.
Collaborator Contribution Our partner in Lisbon has supplied imaging data and other information, as well as interpretation of the results of our wound healing simulation.
Impact The main output has been production of accurate biomechanical models that establish correspondence between prone, supine and upright views of the female breast (Bjoern et al 2016, Han et al 2014) and development a prototype system for predicting breast deformation due to wound healing post lumpectomy (Vavourakis et al 2016).
Start Year 2013
 
Description Multi-scale Computational Anatomy 
Organisation Nagoya University
Country Japan 
Sector Academic/University 
PI Contribution I have provided expertise as a member of the international scientific advisory board to the Japanese Multi-scale Computational Anatomy Programme
Collaborator Contribution I have had access to their technology and research programme which has benefitted several of our programmes, but in particular the Intelligent Imaging EPSRC Programme. In 2015 I was funded by Nagoya on a 3 month sabbatical.
Impact A direct outcome has been a series of papers on image guided gastric surgery (see publications).
Start Year 2015
 
Description Nijmegen 
Organisation Radboud University Nijmegen Medical Center
Department Department of Radiology and Nuclear Medicine
Country Netherlands 
Sector Hospitals 
PI Contribution We established a collaboration with Nijmegen which led to two EU projects HAMAM and PRISM. PRISM arose during the duration of the Programme Grant and the EPSRC MIMIC grant. We contributed expertise and work on image registration and biomechanical modelling.
Collaborator Contribution Nijmegen provide a significant amount of data for both EU projects. They also provided expertise in pathological specimen preparation post mastectomy. Together we developed a method for alignment of histology specimens and medical image data (Mertzanidou et al 2017). Nijmeneg also provided significant expertise in breast imaging and breast cancer screening and diagnosis.
Impact A method for preparation of breast tissue samples for histology and alignment of the resulting data with medicall imaging (Mertzanidou et al 2017). Methods for establishing correspondence between x-ray mammogram breast screening images and MRI (Mertzanidou et al 2014, Mertzanidou et al 2012). A method for characterising stromal maturity and breast cancer risk from H&E stained histological images (Reis et al 2017). This is multi-disciplinary involving computer scientists, physicists, breast pathologists and radiologists.
Start Year 2008
 
Description Philips Research labs Hamburg 
Organisation Koninklijke Philips Electronics N.V.
Department Philips Research Hamburg
Country Germany 
Sector Private 
PI Contribution We have worked with Philips for well over 20 years. For the period covered by the Programme Grant and EPSRC MIMIC we have had two PhD students sponsored by Philips and in-kind support from the team in Hamburg. We have contributed our work and expertise in image registration, image analysis and in particular in biomechanical modelling of breast tissue deformation during the significant manipulations of the breast (e.g. compression) arising during imaging.
Collaborator Contribution Input on the business objectives of Philips. Significant help and expertise in mechanical modelling and image acquisition.
Impact Over the years the output from this collaboration have been substantial and game-changing, in particularly in the area of image registration with some of the most highly cited papers in our field resulting from the joint working (e.g papers by Rueckert, Penney and Studholme between 1996 and 2000). In the time of the Programme grant, MIMIC and the three EU projects HAMAM, MIMIC and in particular PICTURE we have developed innovative technology and published key papers on alignment of X-ray mammograms and MR image (Mertzanidou et al 2012, 2014), produced accurate biomechanical models that establish correspondence between prone, supine and upright views of the female breast (Bjoern et al 2016, Han et al 2014), developed a prototype system for predicting breast deformation due to wound healing post lumpectomy (Vavourakis et al 2016), developed a microstructural model of collagen re-alignment to predict peritumoral stromal changes (Wijeratne et al 2016), developed a system to assess peritumoral stromal maturity in H&E stained histology images of the breast (Reis et al 2017), shown that diffusion MRI can detect stromal anisotropy in breast cancer tissue samples (Bailey et al 2017).
 
Description Roger Bourne 
Organisation University of Sydney
Department School of Pharmacy and Medical Sciences
Country Australia 
Sector Academic/University 
PI Contribution Application of the system developed in Sydney to the prostatectomy specimens collected at UCLH
Collaborator Contribution Access to expertise and experience in developing the system for tissue sample holding and slice preparation of whole prostatectomy specimens.
Impact A methods paper has just been published and the method has been adopted by the prostate group to allow accurate co-registration of clinical and histological imaging.
Start Year 2016