Can we really screen men for prostate cancer with MRI scans? If so, who is going to pay and where will we get all the MRI scanner slots required? Is Artificial Intelligence or AI the ultimate solution?
I was watching the BBC “News at Ten” a few weeks before Christmas when the news of a new prostate cancer screening trial broke. With over 47,000 men diagnosed annually in the UK alone or circa 130 patients per day and an annual mortality rate of around 12,000 I immediately recalled my dad’s treatment pathway for his prostate cancer which was messy and over too quickly to say the least. From a digital examination to biopsy, surgery to a course of external beam radiotherapy and finally treatment for bone mets, meant that he died within a few of years of diagnosis, aged just 69. Even as a “cancer professional”, I felt that with any support I tried to provide I was always playing catch-up and at arms-length if you know what I mean.
Added to the fact that as therapy radiographer and one that ran a business (with very interested financial investors) in prostate cancer treatment with permanent seed implants at its core in the late 90’s I had a strong interest both professionally and from a family perspective. I always wondered whether a proper screening service for men would have made a difference to my dad’s prognosis as I am sure he kept his symptoms to himself for far too long.
The news announced that a team at UCL lead by Prof Mark Emberton plan to ask GP’s in London to randomly gather 300 men aged between 50 and 75 and offer them the chance join a screening trial that will both MRI scan each patient and take a PSA test, combining the results to see whether MRI scans results can replace invasive and “sometimes needless biopsies”.
As part of the project, they also want to see if MRI could also be an effective tool for screening healthy men, in the same way there are national NHS screening programmes to detect signs of breast or cervical cancer.
They then plan to analyse the results using both genomic profiling and machine learning, establishing that MRI can actually replace biopsy concluding that “targeting and treating prostate cancer with advanced imaging would be one of the most disruptive discoveries in men’s health”.
I would argue playing devil’s advocate that this is not as “disruptive” when compared to the advances in MR Linacs and truly adaptive radiotherapy that has been at the forefront of our field for a few years now which has the potential to really disrupt radiotherapy treatment delivery for many cancers and patients alike.
UCL eventually aim to recruit 1,000 men with medium to high risk cancers to find out if MRI can be combined with other high-tech diagnostic tests to predict cancer progression.
The vagaries of PSA tests.
However, all of us working in radiotherapy know the vagaries of the PSA test with its considerable number of false positives leading to needless biopsies and also missing a small but important percentage some higher-grade tumours. However, it is now the policy of NICE who have issued guidance published in May 2019 approving multiparametric MRI (mpMRI) as a first line investigation for men suspected of having clinically localised prostate cancer. This cutting- edge technology can produce a detailed image of the prostate which can help specialists decide whether a biopsy is needed.
What is multiparametric-MRI of the prostate and why do we need it?
There is a paper to read here that summarises the benefits of this form of MRI for Prostate cancer imaging.
My questions relating to this planned screening trial at UCL.
There seems to me to be multiple things being looked at in this proposed screening trial and so these are just a few of the questions that I would like to ask:
1. Can MRI scanning provide a national screening service?
2. Can MRI scanning and PSA filter out the high-risk tumours for urgent treatment?
3. Can MRI combined with PSA tests reduce the need for invasive biopsy and be the sole diagnostic and staging tool?
4. Can MRI scans in combination with other “high-tec” tests predict cancer progression? If so what are they?
5. How will we target the higher-risk cases without screening everyone?
6. Will Artificial Intelligence or AI assist in the process by “screening” the imaging data collected automatically using algorithms that will be co-developed?
Lastly it is also stated that “capturing harmful cancer at the earliest opportunity will lead to quicker treatments, significantly better outcomes for those men affected and ultimately save lives. It will also reduce the numbers of men undergoing biopsy, which in turn will reduce NHS costs and free up staff time”.
As above, while early diagnosis is vitally important so are the advances in the numerous radiotherapy techniques available to these patients (which we will come to later) and while the results of this trial may well reduce the number of biopsies, reduce costs and impact on staff time, what about the flip-side and the impact on MRI services in the UK and are these ready for the huge potential increase in patients being scanned and the requisite image analysis?
We all know that MRI has had a hugely positive impact on prostate cancer diagnosis through much dedicated research already completed but we are no nearer a national screening program as I write. The trial at UCL discusses “targeted” screening with a reduction in biopsy being one end point however there is not a clearly defined criteria for the targeting threshold bar the reference to genomics and machine learning. I have some ideas below.
My recent blogs have focused on AI and machine learning in radiotherapy and it seems to me that screening for prostate cancer with MRI will have to rely on this too in order to cope with the potential sheer volume of data, demand and patient numbers but an automated solution is not simplistic.
Christmas Lectures at the Royal Institution
Not only did the news of Prostate cancer screening grab my attention before Christmas so did the annual Christmas lectures at the Royal Institution…
Started by Michael Faraday in 1825, and now broadcast on national television every year, the Christmas Lectures at the Royal Institution are the UK’s flagship science series. Dr Hannah Fry fronted this year’s lectures and while aimed largely at children and students I couldn’t help but think that part of presentation applied directly to our world!
This is taken from their website:
“We think our lives unfold thanks to a mix of luck and our own personal choices. But that’s not quite true. An unseen layer of mathematics governs every aspect of our world.
Life’s most astonishing miracles can be understood with probability. Big data dictates many of the hot new fashions we follow. Even our choices on Netflix, or our choice of who we marry, is secretly influenced by computer algorithms.
In a series of lectures packed with mind-boggling demos and live experiments, Hannah shows us how to decode life’s hidden numbers; to help us all make better choices, sort fact from fiction, and lead happier lives. But she also warns how our unwavering faith in figures can lead to disaster when we get the sums wrong.
Unravelling suspicious statistics, engineering meltdowns and deadly data, Hannah asks big ethical questions about the trust we place in maths today. Are there any problems maths can’t or shouldn’t solve? Do computer algorithms have too much control over our lives and privacy? Could A.I. decide if someone lives or dies?”
Machine learning or radiologist or both.
I was taken by the “Dog or No Dog” scanner that they invented in one lecture and demonstrated to tell if the image they could see was in fact a dog or not. This becomes even more complex when you want to see if you have a dog or cat! Although the problem sounds simple, it was only effectively addressed in the last few years using deep learning convolutional neural networks. While the dataset is effectively solved, it can be used as the basis for learning and practicing how to develop, evaluate, and use convolutional deep learning neural networks for image classification from scratch and so would also likely need to be applied to a national MRI screening program to cope with the numbers and big data involved. This might in the future be a totally automated results service with radiologist input perhaps at a later stage but with bulk reading of the images performed by AI.
Top 5 Use Cases for Artificial Intelligence in Medical Imaging
Now we believe that it seems MRI screening will have to rely to some extent on AI, I wanted to briefly explore this further. The following recent article looks at the top 5 uses for AI in medical imaging and while the prostate is not mentioned directly, there seems to be scope to create the required algorithms to do this.
The author states: “Artificial intelligence can support radiologists and pathologists as they use medical imaging to diagnose a wide variety of conditions”…
Also that “Artificial intelligence and machine learning have captivated the healthcare industry as these innovative analytics strategies become more accurate and applicable to a variety of tasks.
AI is increasingly helping to uncover hidden insights into clinical decision-making, connect patients with resources for self-management, and extract meaning from previously inaccessible, unstructured data assets.
Medical imaging data is one of the richest sources of information about patients, and often one of the most complex.
With megapixel upon megapixel of data packed into the results from X-rays, CAT scans, MRIs, and other testing modalities, combing through extremely high-resolution images can be challenging even for the most experienced clinical professional.
Artificial intelligence has already proven that it may be a valuable ally for radiologists and pathologists looking to accelerate their productivity and potentially improve their accuracy”.
So how might basic targeting be applied aside from PSA levels?
Prostate Urine Risk (PUR) testing is now being developed…that simply tasks the man to collect the mornings first pee and send it in to the lab. The lab then looks for “gene expression” using prostate bio-markers and so a form of genomics that can distinguish between high/low grade tumours and might hopefully assist in targeted MRI screening?
Research developed by the University of East Anglia and the Norfolk and Norwich University hospital goes someway to achieving this it seems.
This is a synopsis from their publication:
“A new test for prostate cancer that only requires a urine sample, to be collected at home, is being developed by scientists.
Currently there is no single definitive test for prostate cancer, but the NHS says a GP is likely to take a blood sample to look for antigens or perform a rectal exam, or even an MRI scan, which it says can cause “unnecessary anxiety” and stress in men.
The new test, known as the Prostate Urine Risk (PUR) test, being developed by the University of East Anglia and the Norfolk and Norwich University hospital, would “revolutionise” diagnosis.
The at-home collection kit is not only more accessible but also more sensitive than current methods, able to pick up how aggressive the disease is and at what point men will need treatment”
The test would be used for men suspected of having cancer to determine which men have aggressive or intermediate levels of the disease, as well as ruling out those who don’t have it.
It would also be suitable for men already diagnosed with low risk disease and are on active surveillance, known as a “watch and wait” approach.”
It seems that while genetic bio-markers and this test presently have no clinical or public health implications there is a suggestion that it could be a useful screening tool.
Big Numbers for Big Data
There are at my rough estimate circa 9 million men aged 50 to 75 in UK from my basic research and so to screen these numbers annually would have a massive impact on both MR scanner capacity and staffing. Added to the costs involved whereby a basic MR scan might cost the NHS based on recent predictions at least £150 to deliver and a more complex one up to £500 would add by my simple arithmetic a further £1.35 to £4.5 billion to the NHS bill without any further capital investment and additional staff.
By my estimation annual screening would need to find cancer in around 0.5% of the screened population.
MRI and Radiotherapy
In radiotherapy we have become accustomed to the use of MRI imaging in our treatment planning process…
I recall Addenbrookes in Cambridge in the 90’s developing the first basic image fusion package that made quadrants of the CT scan and simply overlaid the MRI image on top. We have progressed a long way from there with the advent of innovative and truly disruptive technology in the form of MR Linacs from Elekta’s Unity to the ViewRay Meridian system.
The prostate is well suited to adaptive radiotherapy as we are aware with many advances in IMRT and VMAT developing directly from treating this disease there is one other I would like to focus on as it was a big part in the development of my company DHA and investment in it.
Trans-perineal, Ultrasound Guided Seed Implants for Prostate Cancer.
In 1996 DHA lost the successful agency business to sell Render Plan 3D in parts of Europe when Elekta purchased the product from Precision Therapy Inc and it became their Precise Plan system. And so while one man’s loss is another man’s gain and this did impact on our business it eventually opened up another huge new market for us in 1997.
A company called Multimedia Medical Systems Inc or MMS wanted to break into Europe with their tongue-twister named product “MMS Therpac Plus 6.6 B3D TUI treatment planning system” and made contact with me. Their new management team were made up by some of my key contacts at Precision Therapy and were keen to continue to use my services.
Brief History of Therpac…
Therpac was the first complete TPS written for the model HP 9830A microcomputer. A Microsoft C based Therpac-PC was introduced in 1986. In 1993, a transperineal ultrasound-guided implant (TUI) module was introduced.
Within a few years, the module could incorporate ultrasound images for pre-planning and CT images for post-implant dosimetry. Multimedia Medical Systems (MMS) Inc were a telemedicine company specializing in video-conferencing and image-sharing. The TUI module with graphical-user interface for the Windows operating system, introduced in 1997.
Collaboration with the physicians and physicists at the Mount Sinai Hospital in New York developed a user-defined dose-optimization module, and the facility for interactive intra-operative planning with real-time display of an implanted dose distribution. Two years later, the MMS program was acquired by Varian Medical Systems, Inc and renamed ‘VariSeed’ it has been the dominant program in use in the US for interstitial brachytherapy treatment planning for over twenty years.
Big in the US but not in Europe
In Europe, while permanent radioactive seed implants were largely a forgotten and historical treatment, the United States proactively treated prostate cancer transperineally using live Ultrasound imaging and Iodine 125 seeds on lots of cases to the extent that in 2004 over 50,000 implants were carried out! In the “States”, patient choice was critical to the extent that men would ask their doctor for this relatively “non-invasive, wake up and your treatment is done” approach based on being an outpatient with interest spread by word of mouth (with no real internet then!) rather than entertain at least 6 weeks of EBRT. These men drove things forward and so we hoped Europe and the UK was ready too.
And, so in 1997 we began to create a business in Europe to market this product and re-establish the trans-perineal technique. We met with all the major seed manufacturers such as Amersham, Bard and Indigo and some leading Urologists such as Dr Frank Kahmann in Mannheim and latterly Berlin and oncologists at leading European centres of excellence such as the Curie Institute in Paris, both whom became users from the start and our early adopters. Within a year we had sold around 20 systems and so were pleased that we had a comprehensive agency contract in place even though Varian had now bought the company. Our success meant that Varian were still keen to work with us, which was good in that I had employed staff to deal with this product directly and while that relationship developed, we sold even more systems in the UK too in both the NHS and especially the private sector who could charge a premium price for a popular, innovative treatment saving on the costs of extensive courses of EBRT.
In 2000 we stopped selling the VariSeed system while Varian took complete control with a very cordial parting of ways.
Permanent Seed Implants
However, the use of permanent seed implants for LDR brachytherapy for early stage prostate cancer increased exponentially. With this technique, radioactive seeds are positioned in the prostate using needles that are guided into position by ultrasound and a template or grid through which the needles are inserted to control their spacing. From 50 to more than 100 individual seeds may be placed in a single session, using 10 to 20 needles. The seeds or radiation sources then deliver the dose slowly over the following month and stay in place for the rest of the patient’s life.
By 2010 there were circa 1,200 hospital sites worldwide using the VariSeed software system for guiding the placement of permanent prostate seed implants. About 60,000, or one third, of U.S. men who were diagnosed annually with prostate cancer had a permanent implant.
The basic technique and program
The VariSeed program ran on a simple laptop PC that would be taken into theatre so that planning could be carried out in real time, a major benefit to all concerned.
In very simple terms the Ultrasound probe was inserted into the rectum and needles inserted via a special grid with the requisite number of seeds preloaded or inserted individually or as part of a chain.
Position of the seeds was recorded in the planning system and MRI scans could be fused to enhance contouring or seed positioning to “active” areas.
The logical tab bar on the top of the VariSeed program allowed simple pre and post planning procedures to be very simply calculated and evaluated. Pre-planning using the Ultrasound Image allowed for real time planning on live images and accurate needle tracking following curved and angled needles and tracking individual seed locations.
Inverse planning based on clinical goals with quality alerts was also possible.
Some key system features included:
The MRI fusion of seed position contained in Ultrasound and CT images with MRI.
An automated “SeedFinder” algorithm allowed the use of a 3-D Digital Reconstructed Radiograph (DRR) generated from a CT to more easily see and interactively add or remove proposed seed locations from the treatment plan.
The ability to contour in the sagittal and coronal planes and for post planning was a useful tool that created a digital reconstruction of the CT image and made it far more simple to define the seed placement, which had been time-consuming in the past.
Sector analysis made it easier to verify seed locations following source localization and reconstruction using the projection display mode.
For those of you wanting to read more on this technique then this is a good overall assessment:
Read about Prostate permanent seed implants here: https://en.wikipedia.org/wiki/Prostate_brachytherapy
Combining MRI and Seed Implants – Screening though to integrated and personalised “tumourlet” treatment.
As we know, Prostate cancer can also be treated with many other treatment modalities…
from HDR Brachytherapy, MR Linac based adaptive radiotherapy, IMRT/VMAT, Surgery, HiFu, Cryotherapy and importantly the option of “watchful waiting” or no treatment at all, all of which benefit from the regular input of data from MRI scans.
The use of spacer gel, injected between the rectal wall and prostate to lift the prostate away from the rectum and reduce rectal dose is also a new advance in regular use.
It has been shown that prostate MRI allows for enhanced contouring of the anatomy of the prostate. MRI/Ultrasound fusion may be useful for improving the accuracy of intraoperative planning as it shows better resolution of the prostate apex. The rectal dose can also be underestimated with CT compared to MRI based pre and post planning. MRI/Ultrasound intraoperative fusion of images can improve prostate dosimetry while sparing the rectum and urethra, potentially impacting disease control and late toxicity.
However, there is now a further advance in the treatment of Prostate cancer with seeds that also uses MRI as an integrated part of the process.
Biopsy mapping…targeted and tracked biopsies…the future?
Historically, the tumour staging biopsy for Prostate cancer uses a “trans-rectal” technique whereby the samples taken or cores would be analysed, reported on but that critical information regarding the precise location of that core, was essentially thrown away!
A leading Urologist who uses a form of Robotic Prostatectomy and performs seed implants has developed an alternative biopsy technique whereby the biopsy uses the trans-perineal technique with Ultrasound and a needle template (the same equipment that delivers the seed implants above) so that each core taken can be assessed, the Gleeson stage calculated and the results of this fed back into the VariSeed system so that informed choices can be made regarding the most suitable treatment option of the patient.
In the case of seed implants the treatment plan and seed location can be weighted to allow a more focused radiation dose where the cancer is most active, or by using Dicom RT can transfer the data to other treatment planning systems for HDR Brachytherapy offering a dominant lesion boost, to a Linac delivering IMRT with a benefit of reduced treatment side effects or even to HiFu systems, Cryotherapy and also to assist in any decision on the option and ongoing management of “watchful waiting”.
Subsequent post treatment or watchful waiting biopsies can follow an identical process and so is also a very useful follow up tool.
A further moduleof VariSeed that allows the merging of CT, PET (usually in recurrent cases), MRI and Ultrasound images can superimpose the Ultrasound template as with seed implant planning but allow visualisation of each core biopsy location and then following the results of histology input the staging into a 3D image showing the location of active “Tumourlets” that require additional attention or targeted treatment. A Radiologist will have marked up the tumour and critical structures previously and these would become part of a Dicom compliant data set for sharing with any treatment modality, fused with any diagnostic images allowing for a completely bespoke treatment.
…and as I write!
Promising initial results show how AI may improve breast cancer screening and ease pressure on the NHS…in association with Google Health…
As I write this the day following New Year’s Day it has just been announced on the radio I am listening to and on Sky TV later that:
Artificial Intelligence better than expert radiologists at spotting breast cancer, study says
Cancer Research UK says “promising” initial results show how AI may improve breast cancer screening and ease pressure on the NHS.
An artificial intelligence programme has been developed that outperforms clinical experts in detecting breast cancer in mammograms, according to researchers.
The study found that an AI system developed by Google Health can identify cancer in breast screening mammograms with fewer false positives and fewer false negatives than radiologists.
And so that is good place to end my blog looking at the use of MRI in screening and some key radiotherapy treatment advances in Prostate cancer enhanced by MRI.
Duncan Hynd – January 2020