Case study – evaluation of Pi™ with GE Healthcare and EMRAD
MRI and prostate cancer diagnosis
Over the last 10 years, MRI has been introduced worldwide to help diagnose prostate cancer. Many patients with prostate cancer are over-diagnosed and over-treated, and many others are under-diagnosed and under-treated. MRI is helping to address these issues and is reshaping prostate cancer diagnosis and treatment . MRI is key because it is widely available, safe, low-cost, and can help patients avoid a painful biopsy that can have serious side-effects. It is the only imaging test that can reliably detect prostate cancer and differentiate it from benign conditions.
Improving the accuracy of cancer detection using MRI and Pi™
Lucida Medical was founded in 2019 by Prof Evis Sala and Dr Antony Rix with the vision to improve the accuracy, cost-effectiveness and productivity of screening for cancer using MRI. The company’s first product, Pi™ is a CE-marked medical device software that uses AI to assist prostate cancer diagnosis.
A patient with suspected prostate cancer has tests and a pre-biopsy MRI as normal. Pi™ then produces images identifying the prostate and suspicious lesions, and a template report with risk scores. Together, these are intended to assist the radiologist to locate cancer or rule out individuals who can avoid biopsy. A further potential benefit is that this automated analysis could save radiologists’ time.
How can we integrate the benefits of AI into radiologists’ workflow?
A study using historical data indicates that Pi™ has 93% sensitivity to significant prostate cancer at 24% false positive rate . If this performance can be matched when it is used in hospitals, Pi™ could help reduce missed cancers and avoidable biopsies. But for this to be possible, it is essential that the software is easy for radiologists to use.
Collaboration with GE Healthcare and EMRAD
In 2021 Lucida Medical was selected from 350 applicants to be one of 6 companies included in GE Healthcare’s Edison™ Accelerator programme. Lucida Medical worked with GE Healthcare and the East Midlands Imaging Network (EMRAD), a pioneering organisation which has pooled radiology resources amongst a group of eight NHS trusts from across the East Midlands. The goal of this service evaluation project was to understand whether GE Healthcare’s Edison™ platform could enable Pi™ to easily fit in with the radiologists’ workflow.
Lucida Medical and GE Healthcare installed Pi™ in EMRAD’s data centre. GE Healthcare’s Edison™ Open AI Orchestrator provided the link between Pi™ and the radiology information system (RIS), picture archiving and communications system (PACS) and viewer, and reporting software used by hospital teams. With data remaining inside EMRAD’s secure firewall, example anonymised prostate MRI studies were processed automatically by Pi™ and the outputs were made available for radiologists to review alongside the original MRI images.
How the system works
illustrated with an example prostate MRI study [2,5,6].
“Quality of software is impressive. The most suspicious lesions are marked in red for easy and quick visualisation and implemented in the body of report according to level of conspicuity.”
Summary of findings
The evaluation of this installation led to the following conclusions:
- The radiologists reported no disruption to their workflow and were satisfied with the overall experience
- Experience with the AI outputs was positive, with the most suspicious lesions highlighted for radiologists to review, and with volume measurements that could save time.
- Cancer Research UK, Prostate Cancer Statistics, https://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/prostate-cancer
- Clark K, et al. The Cancer Imaging Archive (TCIA): Maintaining and Operating a Public Information Repository, J Dig Im, 26(6), 1045-1057, 2013. https://doi.org/10.1007/s10278-013-9622-7
- Drost FJH, et al. Cochrane Sys Rev Diag. 2019 https://doi.org/10.1002/14651858.CD012663.pub2
- International Agency for Research on Cancer, https://gco.iarc.fr/today/online-analysis-table?v=2020
- Litjens G, et al. Computer-aided detection of prostate cancer in MRI, IEEE Trans Med Im 2014;33:1083-1092. https://doi.org/10.1109/TMI.2014.2303821
- Litjens G, et al. ProstateX Challenge data, The Cancer Imaging Archive (2017). https://doi.org/10.7937/K9TCIA.2017.MURS5CL
- Lucida Medical estimates based on published studies and the NPCA Annual Report 
- NICE Guideline, Prostate cancer: diagnosis and management, May 2019 https://www.nice.org.uk/guidance/ng131
- National Prostate Cancer Audit, Results of the NPCA Prospective Audit in England and Wales for men diagnosed from 1 April 2018 to 31 March 2019, January 2021 https://www.npca.org.uk/reports/
- Royal College of Radiologists, 2020 UK workforce census https://www.rcr.ac.uk/system/files/publication/field_publication_files/clinical-radiology-uk-workforce-census-2020-report.pdf
- Suchanek J, et al. Multi-stage AI analysis system to support prostate cancer diagnostic imaging. EuSoMII Virtual Annual Meeting, 24 October 2020, https://doi.org/10.26226/morressier.5f7f3e3d6934880e60c0a8b1
The figure above illustrates the RTSTRUCT format segmentations output by the software. The segmentations are overlaid, on a 3D multi-planar reconstruction of the T2 axial image, together with the 3D mesh view.
Colours shown are:
• Red: index lesion
• Blue: prostate organ
• Grey: seminal vesicles
Smaller lesions are also visible, with the smallest indicated by the software using labels (e.g. Lesion 5-1).
The link below is to a video presentation by Professor Richard Hindley the lead Consultant Urological Surgeon for the study at HHNFT and Dr Aarti Shah, the lead Consultant Radiologist for the study at HHNFT
Managing Prostate Cancer with MRI and Pi by Dr. Aarti Shah & Prof. Richard Hindley
PAIR-1 Retrospective cohort validation study of AI-based prostate cancer diagnosis support software Pi ™ led by Hampshire Hospitals NHS Foundation Trust (HHNFT) and Lucida Medical
Why do we need this research?
Prostate cancer is the most common cancer among men in the UK. It is estimated that 120,000 men go through diagnosis, 47,000 are diagnosed with prostate cancer, and 11,000 die from the disease each year. Magnetic resonance imaging (MRI) scans have recently been introduced at the very beginning of the patient pathway to help reduce biopsy rates and improve detection of significant cancers that require treatment.
The interpretation of prostate MRI scans requires time and expertise, leading to variability in reporting capacity and accuracy as well as hindering adoption of pre-treatment MRI across the NHS. This has led to interest in the potential to apply artificial intelligence (AI) computer software to improve the process.
Lucida Medical has developed AI software Pi, to support prostate cancer diagnosis. Preliminary results suggest that the software performs well compared to radiologists and alternative systems. In this collaborative project, we will check and if necessary adjust (calibrate) its settings, validate its expected performance, and test it.
We anticipate Pi will detect significant cancers with equal or better performance than world-leading hospitals. Consequently, it could reduce healthcare inequality, save the NHS money, and give patients confident early diagnosis.
To make this work and to provide evidence on the potential performance improvements, we need to obtain more data from representative prostate cancer patients who have been through the process of diagnosis in NHS hospitals.
For this study, we plan to access the records of up to 2100 patients who previously went through prostate cancer diagnosis. Data from these patients would be de-identified and there will be no impact on their care.
Just over half of this data will be used first to make sure that the software will work well across the full range of scanners and scanning methods in different hospitals. The developers will use this data to check that the AI software is correctly calibrated, and to change settings if necessary. The remainder of the data will then be used to test how well the AI software works with unseen patients, paving the way for it to be used in a clinical setting.
Prostate cancer patients and relatives have helped plan this study. Working with them ensures this research is acceptable and provides vital patient input into the design of this work and how we communicate with patients and the wider public.
We will publish the results to give doctors and patients confidence that the AI system can be used for accurate prostate cancer diagnosis. This study will help demonstrate the accuracy of the AI system for use in hospitals across the NHS.
Papers and conference presentations reporting the results of the PAIR-1 study are listed here.
We collaborate with Cambridge University, we are a partner in the ReIMAGINE prostate imaging trial, are in the process of establishing collaborative partnerships with several UK NHS Trust hospitals to help further develop and validate and test Pi, our AI system for detecting prostate cancer from MRI.