Personalised lung cancer treatment through outcomes predictions and patient stratification
Lead Research Organisation:
University College London
Department Name: Medical Physics and Biomedical Eng
Abstract
In this fellowship, I will use a radical new approach to improve the radiotherapy treatment of patients suffering from inoperable non-small cell lung cancer (NSCLC). NSCLC is a cancer of unmet need for which the actual chemo-radiotherapy treatment has remained mostly unchanged for more than 30 years, with a poor 16.4% 5-year survival. This poor survival is caused by the limitation of the 'one-dose-fits-all' paradigm which neglects the diverse spectrum of clinical presentation in NSCLC. To improve the treatment, my group and I will harness the capacities of novel cutting-edge artificial intelligence techniques combined with a massive retrospective database of patients data to answer a fundamental question about lung cancer which is "How will the disease progress?". More precisely, the deep learning approach will be used to extract general trends relating patient's data features (histopathology, anatomy, tumour stage, tumour activity, treatment plan) to an outcome (death, recurrence, secondary fibrosis, heart failure and success). The methodology output will then be used for two endpoints of the study. It will first be directly used to inform and personalise the radiotherapy treatment planning strategy to improve patient survival. It will also serve as a basis to define a new stratification procedure for lung cancer patients to refine the clinical trials selection system. This framework will enact a paradigm change in treatment planning for radiotherapy and has the potential to enable a jump in performance of the treatment by tailoring the dose to the patient; thereby lowering the secondary effects and improving overall survival.
Planned Impact
The primary beneficiaries of this project are people suffering from inoperable stage III NSCLC, through the hypothesised jump in survival enabled by the personalised therapy. Specific steps and objectives have been outlined in this fellowship to reach the patient which includes the development of a robust Code of Practice and a clinical trial. Furthermore, by directly involving patients, clinicians, and end-users in the development of the framework and the design of the clinical trial, I will ensure that the framework reaches the beneficiaries.
Additional beneficiaries include 1) academics in the fields of medical physics, radiotherapy and oncology who will benefit from the research advance in each of these specific fields. High-impact open-access publication and conference presentation will be sought in the field of diffeomorphism, machine learning, and personalised dosimetry. An international workshop will be organised to disseminate our results and to gather experts in the field who will provide critical feedback. Although every aspect of this projects is covered by the established collaborations, I will work with any interested new collaborators, as long as appropriate protection of the intellectual property on the developed framework is defined, which I will seek in collaboration with the UCL Business office.
Health services will benefit from the reduced cost of treatment, and both clinicians and carers will benefit from the increased efficiency of the treatment. To ensure that the technique developed aligns with the need and tolerance of the patients and carers, I will circulate the ideas and consult a patient and public involvement group, who will be selected with the UCL Medical Research Council Clinical Trials Units PPI Groups, that will help transition the framework to enable personalisation and collaboration in care decision making agenda. This group will include patients, carers and the general public interested in radiation oncology who will provide feedback on the framework, including advice on the research idea, study design, developing accessible information sheets and dissemination.
Additional beneficiaries include 1) academics in the fields of medical physics, radiotherapy and oncology who will benefit from the research advance in each of these specific fields. High-impact open-access publication and conference presentation will be sought in the field of diffeomorphism, machine learning, and personalised dosimetry. An international workshop will be organised to disseminate our results and to gather experts in the field who will provide critical feedback. Although every aspect of this projects is covered by the established collaborations, I will work with any interested new collaborators, as long as appropriate protection of the intellectual property on the developed framework is defined, which I will seek in collaboration with the UCL Business office.
Health services will benefit from the reduced cost of treatment, and both clinicians and carers will benefit from the increased efficiency of the treatment. To ensure that the technique developed aligns with the need and tolerance of the patients and carers, I will circulate the ideas and consult a patient and public involvement group, who will be selected with the UCL Medical Research Council Clinical Trials Units PPI Groups, that will help transition the framework to enable personalisation and collaboration in care decision making agenda. This group will include patients, carers and the general public interested in radiation oncology who will provide feedback on the framework, including advice on the research idea, study design, developing accessible information sheets and dissemination.
Organisations
- University College London (Fellow, Lead Research Organisation, Project Partner)
- University College London (Collaboration)
- Magna Græcia University (Collaboration)
- Royal Free London NHS Foundation Trust (Collaboration)
- Polytechnique Montréal (Project Partner)
- University of Montreal (Project Partner)
- National Physical Laboratory (Project Partner)
- Centre Hospitalier de l’Université de Montréal (Project Partner)
People |
ORCID iD |
Charles-Antoine Collins-Fekete (Principal Investigator / Fellow) |
Publications
Bär E
(2021)
Assessment of the impact of CT calibration procedures for proton therapy planning on pediatric treatments.
in Medical physics
Fullarton R
(2023)
A likelihood-based particle imaging filter using prior information.
in Medical physics
Robertson D
(2023)
High-Density Scintillating Glasses for Integrating-mode Particle Radiography
Shen Z
(2023)
MO-0642 Predicting post-radiotherapy cardiac toxicity for non-small cell lung cancer patients
in Radiotherapy and Oncology
Shen Z
(2023)
Modeling Acute Chemoradiotherapy (CRT) Diarrhea Severity Using Automatically Contoured Small Bowel
in International Journal of Radiation Oncology*Biology*Physics
Simard M
(2022)
MO-0216 Integrated-mode proton radiography using 2D lateral projections and a scintillator
in Radiotherapy and Oncology
Description | One of our pathology algorithms has a significantly high capacity of producing sarcoma diagnosis, with an accuracy of 88% on one diagnosis, and a accuracy of 99% if three diagnosis are involved. We are currently in the process of seeking funding to evaluate it against pathologists in 15 NHS Trusts. |
Exploitation Route | After extensive validation against pathologists, we will seek industrial partners to help us bring this new AI algorithm to clinical usage. |
Sectors | Healthcare |
Description | We are currently with a portion of my team looking into the commercialisation of some of the Digital AI pathology algorithm we developped. |
First Year Of Impact | 2024 |
Sector | Healthcare |
Impact Types | Economic |
Description | Fourth ion imaging workshop 2023 in London, UK |
Geographic Reach | Multiple continents/international |
Policy Influence Type | Contribution to new or improved professional practice |
URL | https://ionimaging2023.sciencesconf.org/ |
Description | AI-based diagnosis for improving classification of bone and soft tissue tumours across the UK |
Amount | £613,172 (GBP) |
Funding ID | EP/Y020030/1 |
Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
Sector | Public |
Country | United Kingdom |
Start | 09/2023 |
End | 04/2025 |
Description | Algorithmic development of proton radiography for image-guided proton radiotherapy of lung cancer |
Amount | € 224,933 (EUR) |
Funding ID | H2020-MSCA-IF-2020 101023220 |
Organisation | Marie Sklodowska-Curie Actions |
Sector | Charity/Non Profit |
Country | Global |
Start | 08/2021 |
End | 10/2023 |
Description | An artificial intelligence framework to classify soft tissue tumours |
Amount | £50,000 (GBP) |
Funding ID | SUK18.2021: |
Organisation | Sarcoma UK |
Sector | Charity/Non Profit |
Country | United Kingdom |
Start | 03/2022 |
End | 04/2023 |
Description | Computational pathology to risk stratify solitary fibrous tumours (SFT): an integrated approach using digital pathology, machine learning and genetics. |
Amount | £40,000 (GBP) |
Organisation | Pathological Society |
Sector | Charity/Non Profit |
Country | United Kingdom |
Start | 08/2021 |
End | 02/2022 |
Description | EPSRC Doctoral Training Partnerships Funding |
Amount | £140,000 (GBP) |
Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
Sector | Public |
Country | United Kingdom |
Start | 08/2023 |
End | 09/2027 |
Description | Heavy Ion Therapy Research Integration - Research access |
Amount | € 8,000 (EUR) |
Organisation | GSI |
Sector | Private |
Start | 02/2024 |
End | 12/2024 |
Description | Personalised lung cancer treatment through outcomes predictions and patient stratification |
Amount | £1,149,959 (GBP) |
Funding ID | MR/T040785/1 |
Organisation | United Kingdom Research and Innovation |
Sector | Public |
Country | United Kingdom |
Start | 08/2020 |
End | 09/2024 |
Description | Proton Radiography For Adaptive Radiotherapy And Image Guidance Using A Novel Image Detection System |
Amount | £70,153,490 (GBP) |
Funding ID | NIHR 205508 |
Organisation | National Institute for Health Research |
Sector | Public |
Country | United Kingdom |
Start | 12/2023 |
End | 12/2026 |
Title | - Organized AI Framework for Radiology Outcome Prediction, Classification and Object Detection (2021) |
Description | Organized AI framework that allows inputting a variety of modalities from an online catalogue (XNAT platform hosted at UCL) and perform AI training, using state of the art models, to correlate outcomes to a variety of clinical and imaging factors. These framework will underpin all activities in our future research in AI for radiology, and will be the building block towards many high-impact publication. |
Type Of Material | Improvements to research infrastructure |
Year Produced | 2021 |
Provided To Others? | Yes |
Impact | Uniformization of AI research at UCL for many groups under the same framework, direct access through the XNAT platform to the clinician that can throw AI jobs without requiring deep knowledge of programmation. Building block for future research. |
URL | https://github.com/cacof1/OutcomePrediction |
Title | Organized AI Framework for Pathology Outcome Prediction, Classification and Object Detection |
Description | In collaboration with the UCL Department of Pathology, we developed and object analysis, classification and prognostic AI model to be used on whole slide images. The model will be used for analysis on both the sarcoma aspect of research (in collaboration with Prof. Adrienne Flannagan), as well as with the Non-Small cell Lung Cancer Biopsy currently acquired to help improve prognostic on outcome. The model uses multi-zoom and multi-model input to enable state-of-the-art prediction, and is equipped with the latest development in image preprocessing for pathology. |
Type Of Material | Improvements to research infrastructure |
Year Produced | 2021 |
Provided To Others? | Yes |
Impact | The model is already being used in the development of two manuscripts: - Mitosis Detection for Solitary Fibrous Tumour using Anti-phosphorylated Histone H3 (pHH3) Antibodies and Digital Pathology - Predicting soft tissue sarcoma diagnosis with deep learning And will be the basis for pathology analysis of our future work. |
URL | https://github.com/cacof1/DigitalPathologyAI |
Title | Web Application to deploy easily our AI model |
Description | We launched a web application that enables direct use of our AI model for digital pathology without any knowledge about progamming. This platform is aimed at making it easy for clinicians and researchers to access and work with our AI tools for analyzing pathology slides. In line with our commitment to open science, we've also open-sourced the AI model and published detailed documentation on its development and performance. This approach is designed to encourage collaboration and facilitate the model's adoption and improvement by the global scientific and medical communities. We hope to drive engagement of the clinican and that the model can be used in a clinical trial scenario to provide proof for certification. |
Type Of Material | Improvements to research infrastructure |
Year Produced | 2023 |
Provided To Others? | Yes |
Impact | We are already collaborating with engaged clinician at Sheffield Hospital, the Royal National Orthopaedic Hospital, and at the School of Medicine, Medical Sciences & Nutrition from the University of Aberdeen that are testing our models for sarcoma classification, lymphocites detection and mitoses detection. |
URL | http://www.sarcoma-ai.com |
Title | MODAL: Multi-Omics Data Lake |
Description | Dataset collection of Cancer Patient including: - Anatomical images (CT, CBCT, MRI) - Functional Imaging (PET) - Clinical Features The dataset contains 2621 subjects, and 4475 imaging sessions across 12 projects: Rectal (UCLH, 20 patients, RADIANT/Aristotle clinical trial, 596 patients, Darius (n=20)) Lung (RTOG-0617 (n=476), National Lung Screening Trial (n=265), IDEAL (n=57), NSCLC Radiomics (n=422), Lung_Seg (n=60), NSCLC Radiogenomics (n=200), UCLH (n=451)) Liver (UCLH, (n=63) Esophageal (UCLH (n=98) |
Type Of Material | Database/Collection of data |
Year Produced | 2022 |
Provided To Others? | No |
Impact | This unified database creation request received Approval from the HRA to start collecting a large amount of fully anonymised patient data to be used for AI training and inference. This large scales dataset encompass multiple cancer and multiple clinical trials. |
Title | OctoPath: Large Scale Pathology Single-Cell Database from Immuno-Fluorescence cells |
Description | Through the course of the UKRI AI for Health Award, we generated multi-millions single-cell labels for AI training of various cells (mitotic cells, epithelial, lymphocytes CD4/CD8, stromal, muscular, endothelial, macrophages). This database of single cells contains the H&E representation, the mask around the nuclei, as well as the label for each cell. It will be of great usage for any future AI algorithms that wants to train on these datasets and produce new AI algorithm. The database is currently compiled and will soon be published. |
Type Of Material | Database/Collection of data |
Year Produced | 2024 |
Provided To Others? | Yes |
Impact | Our current publication submitted to Nature: Machine Learning come directly from the generation of this database. As it is yet to be made fully public (last validation), no external impacts have yet been recorded. This section will be filled |
Title | UK-Based Sarcoma WSI Database |
Description | Before 2023: Database of 10k scanned sarcoma whole-slide images, of 22 different diagnosis. This database is not yet public as we are waiting for our own publications to come out, as well as for ethical regulatory approval. Since 2023 with the award of the UKRI AI for Health, we have increased our current number of diagnosis to 29 and currently are collecting 35,000 WSI to train our AI model. |
Type Of Material | Database/Collection of data |
Year Produced | 2023 |
Provided To Others? | No |
Impact | The large scale database has allowed us to significantly improves the results of our AI algorithms and to produce state-of-the-art results that are incorporated into our latest model. This model will be tested against pathologist and, pending positive results, incorporated in clinical practice through industrial partners. |
Description | Data Collection - Royal Free Hospital |
Organisation | Royal Free London NHS Foundation Trust |
Country | United Kingdom |
Sector | Public |
PI Contribution | Establish an HRA-approved collaboration with the Royal Free hospital to collect non-small cell lung cancer data as well as thoracic cancer and liver cancer. |
Collaborator Contribution | They are providing in-kind clinical fellow time to help identify patients that are relevant to our data-study, to be added to our extensive database to improve the AI model outputs. |
Impact | No outcomes yet as we are collecting, curating, and aggregating in the database the new cohort of patients data. |
Start Year | 2022 |
Description | Radiology/Pathology AI Outcome Prediction |
Organisation | Magna Græcia University |
Country | Italy |
Sector | Academic/University |
PI Contribution | This collaboration aims at developing a partnership and framework for the prediction of outcome and toxicity in lung cancer. It includes expertise from UCL (Department of Medical Physics - Radiology, Cancer Institute - Histopathology data) as well as the University of Magna Graecia for image manipulation and analysis algorithms. Our contribution is to lead the development of the AI framework for collaboration and to manage the expected outcome and publication between each partners. We are spearheading the AI software development. UCL Medical Physics is mostly developing the framework for Radiology and Pathology (see Methodology section). We also led the ethical aspect of the project for data acquisition in the hospital, including a data request from the Health Research Authority. |
Collaborator Contribution | The contribution of the UCL Cancer Institute is to provide domain knowledge and images for histopathology and to help develop and guide the AI algorithms. They have provided several pathologist trainees to help inject domain knowledge in the form of hand-drawn annotation, and we have co-led grants together in the development of AI for histopathology. The contribution of the University of Magna Graecia is to provide image regularization software for radiology, which will help us minimize data difformation between patients and improve accuracy. |
Impact | Sarcoma UK Grant Jean Anderson Pathology Society Grant Many publications under preparation |
Start Year | 2021 |
Description | Radiology/Pathology AI Outcome Prediction |
Organisation | University College London |
Department | UCL Cancer Institute |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | This collaboration aims at developing a partnership and framework for the prediction of outcome and toxicity in lung cancer. It includes expertise from UCL (Department of Medical Physics - Radiology, Cancer Institute - Histopathology data) as well as the University of Magna Graecia for image manipulation and analysis algorithms. Our contribution is to lead the development of the AI framework for collaboration and to manage the expected outcome and publication between each partners. We are spearheading the AI software development. UCL Medical Physics is mostly developing the framework for Radiology and Pathology (see Methodology section). We also led the ethical aspect of the project for data acquisition in the hospital, including a data request from the Health Research Authority. |
Collaborator Contribution | The contribution of the UCL Cancer Institute is to provide domain knowledge and images for histopathology and to help develop and guide the AI algorithms. They have provided several pathologist trainees to help inject domain knowledge in the form of hand-drawn annotation, and we have co-led grants together in the development of AI for histopathology. The contribution of the University of Magna Graecia is to provide image regularization software for radiology, which will help us minimize data difformation between patients and improve accuracy. |
Impact | Sarcoma UK Grant Jean Anderson Pathology Society Grant Many publications under preparation |
Start Year | 2021 |
Description | Third ion imaging workshop 2022 in Munich, Germany |
Form Of Engagement Activity | A formal working group, expert panel or dialogue |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | The 3rd Ion Imaging Workshop took place on October 13th and 14th at the LMU main building in Munich and was following workshops in Lyon in 2018 and Manchester in 2019. The workshop attracted 50 registered participants, which marked an increased participation compared to Lyon (36 participants) and Manchester (42 participants). The workshop had an international character, with participants from Norway, France, Italy, Germany, Switzerland, the United Kingdom, the Netherlands, Austria, and the United States. External funding was instrumental in ensuring a well-attended workshop, allowing the organizing committee to invite speakers. |
Year(s) Of Engagement Activity | 2022 |
URL | https://ionimaging.org/ws2022munich-summary/ |