AI-driven companion system for prostate cancer management in patients on active surveillance

Award details

AIM OF THE PROJECT

Develop an AI-based system for automated MRI-based prostate cancer (PCa) progression assessment for patients enrolled in active surveillance (AS). The system will build upon AutoProstate, a deep learning-based computer-aided detection (CAD) system prototype which detects significant prostate lesions using MRI.

The specific objectives of this project are:

Objective 1: Extend AutoProstate to work in active surveillance (AS) setting. Patients enrolled in AS usually have early-stage tumours that are more difficult to detect and characterize in terms of PIRADS assessment. CAD systems performing well on aggressive and high-risk PCa may fail to detect low-risk lesions. For this reason, AutoProstate will need to be adapted and tuned to the new specific problem.

Objective 2: Develop a new component to assess longitudinal changes in clinical and imaging data. This component will monitor the PCa progression using clinical and imaging-derived parameters to decide whether the patient should remain in AS or switch to active therapy by applying novel deep learning-based approaches such as multimodal Recurrent Neural Network and Transformer.

Objective 3: Generate a standardized report. The report will support and guide clinicians in the management and long-term follow-up of patients in AS settings.

 

PROJECT DESCRIPTION

Prostate cancer (PCa) is the most frequently diagnosed cancer in men in the UK and the second leading cause of cancer death, with an estimated 52,300 new diagnoses (2016-1018) and 12,000 deaths (2017-2019) every year[1]. Active surveillance (AS) is a conservative management protocol for prostate cancer offered to low- and intermediate-risk patients, which avoids long-term adverse effects on the patient's quality of life[2]. Current AS protocols are based on regular clinical examinations, prostate specific antigen (PSA) testing, and biopsy. Prostate biopsies can cause complications such as infection, bleeding, and urinary retention[3]. In this context, there has been an increased interest in using multiparametric/biparametric magnetic resonance imaging (mp/bp-MRI) alongside PSA for non-invasively monitoring patients in AS[4]. The combined use of MRI technology with artificial intelligence (AI) could represent a shift in how AS is currently performed. However, achieving this ambition is currently challenged by several critical aspects, such as biomarkers definition and selection. Early detection of disease progression using mp/bp-MRI remains suboptimal as there are no specific biomarkers or clear definitions of what constitutes a ‘significant’ change for patients to transfer to further treatment.

This PhD project aims to develop an AI-based system for automated MRI-based PCa detection and progression assessment. The project will make use of a cohort of AS patients collected at Guy's and St Thomas' hospitals with a minimum of three timepoints per patient.

The project outcome will be a pipeline of 3 main components: i) lesion detection and PIRADS assessment, ii) lesion progression prediction, ii) clinical report generation. Each component will be developed as follows:

WP1 – Lesion detection and PIRADS assessment(Y1)

Within the AS regimens, patients usually have earlier-staged and low-risk tumours that tend to represent the more challenging cases for the radiologists to characterize in terms of PIRADS score. The PhD candidate will extend and tune the deep-learning-based CAD system AutoProstate[5] to work in AS settings. AutoProstate will be extended to detect also non significant lesions (PIRADS=2 and 3). New approaches such as nn-Unet V2 [6][7] or Segment Anything Model (SAM)[8]. Moreover, new functionality will be added to assign the PIRADS score to each lesion.

WP2 – Lesion progression prediction(Y2-Y3)

A novel component will be added to AutoProstate to assess longitudinal changes in clinical and imaging-derived data, monitor disease progression, and decide whether to remain in AS or switch to active therapy. Existing regression-based models do not consider temporal dependencies, require the alignment of patients' trajectories, and rely on independent biomarker modelling. To overcome these limitations, deep learning-based models such as recurrent neural networks (RNNs)[9] and Transformer[10] will be explored. The student will also investigate other approaches proposed in Computer Vision for video prediction and not applied to the field of medical imaging [11]. The component will use a variety of features as inputs, including clinical features, lesion-level biomarkers such as shape, volume and radiomics features, and features extracted from the whole prostate anatomy. Since longitudinal cohorts are not acquired in a standardized way and can contain an arbitrary number of timepoints and missing biomarker values, Ad-Hoc training strategies to handle missing data in RNNs will be developed.

WP3- Clinical report generation(Y3)

There is high variability in clinicians' reporting. To report a patient at a specific timepoint, radiologists usually inspect all past scans and read past reports (which may have been reported by different doctors or acquired from different hospitals). The system will automatically generate a standard report following the PRECISE guidelines to facilitate and speed up AS reporting and avoid misunderstanding of intent of reports among radiologists.  

References

1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F, Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. Cancer J. Clin. 2021, 71, 209–249.
2. Garisto JD, Klotz L, Active Surveillance for Prostate Cancer: How to Do It Right. Oncology (Williston Park). 2017, 15, 31(5):333-40.
3. Loeb S, Vellekoop A, et al., Systematic Review of Complications of Prostate Biopsy, European Urology, Volume 64, Issue 6, 2013, 876-892.
4. Briganti A, Fossati N, et al., Active Surveillance for Low-risk Prostate Cancer: The European Association of Urology Position in 2018, Europen Urology, (2018), 357-368.
5. Mehta P., Antonelli M, Singh S, Grondecka N, Johnston EW, Ahmed HU, Emberton M, Punwani S, Ourselin S, AutoProstate: Towards Automated Reporting of Prostate MRI for Prostate Cancer Assessment using Deep Learning Cancer 2021 13(23):6138.
6. Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature methods,(2021) 18(2), 203-211.
7. https://github.com/MIC-DKFZ/nnUNet
8. Kirillov A et al., Segment Anything, arXiv 2304.02643.
9. Nguyen M et al., Predicting Alzheimer’s disease progression using deep recurrent neural networks, Neuroimage, (2020), 222:117203
10. Hu Z., et al., VGG-TSwinformer: Transformer-based deep learning model for early Alzheimer’s disease prediction, Computer Methods and Programs in Biomedicine (2023), 229:107291.
11. Oprea s., A Review on Deep Learning Techniques for Video Prediction, IEEE Transactions on Pattern Analysis and Machine Intelligence, (2022) 44:2806-2826.
12. Moore CM., et al., Reporting Magnetic Resonance Imaging in Men on Active Surveillance for Prostate Cancer: The PRECISE Recommendations—A Report of a European School of Oncology Task Force, European Urology, (2017), 71:648-655.

Informal email enquiries from interested students to the supervisor are encouraged (contact details below).

 Dr Michela Antonelli - email: michela.antonelli@kcl.ac.uk

 

Award value

The studentship is fully funded for 3.5 years. This includes home tuition fees, stipend and generous project consumables.

Stipend: Students will receive a tax-free stipend at the UKRI rate of £20,622 (AY 2023/24) per year as a living allowance.

Research Training Support Grant (RTSG): A generous project allowance will be provided for research consumables and for attending UK and international conferences.

Eligibility criteria

Candidates who meet the eligibility requirements for Home Fee status will be eligible to apply for this project. Home students will be eligible for a full UKRI award, including fees and stipend, if they satisfy the UKRI criteria below, including residency requirements. To be classed as a Home student, candidates must meet the following criteria:

• be a UK National (meeting residency requirements), or
• have settled status, or
• have pre-settled status (meeting residency requirements), or
• have indefinite leave to remain or enter.

Prospective candidates should have a 1st or 2:1 M-level qualification in Biomedical Engineering, Physics, Engineering, Computer Science, Mathematics, or a related programme.

Preference will be given to candidates with a background conducive to multidisciplinary research and preferably programming skills.

We welcome eligible applicants from any personal background, who are pleased to join diverse and friendly research groups.

Application process

Please submit an application for the Biomedical Engineering and Imaging Science Research MPhil/PhD (Full-time) programme using the King’s Apply system. Please include the following with your application:

• A PDF copy of your CV should be uploaded to the Employment History section.
• A 500-word personal statement outlining your motivation for undertaking postgraduate research should be uploaded to the Supporting statement section.

Funding information: Please choose Option 5 “I am applying for a funding award or scholarship administered by King’s College London” and under “Award Scheme Code or Name” enter BMEIS_DTP_MA. Failing to include this code might result in you not being considered for this funding.

The closing date is 12th June 2023 (please note that the applications can be closed earlier if a suitable candidate is found).