An Image is Worth a Thousand Blood Tests
Medical diagnostics is a complex and ever changing beast. New techniques and technologies enable more accurate and personalised diagnosis. Alongside this, recent advances in artificial intelligence are poised to change the way we diagnose and treat patients. And yet, despite the growing availability of technologies, the medical community has maintained its historically slow pace of change.
With prostate cancer, we still see tests like prostate specific antigen (PSA), a blood test, uses as a primary diagnostic tool. The problem with the test, however, is it has a false positive rate of 50% on the conservative side. Clinicians know this and follow up tests like biopsies can confirm these diagnoses but come at high cost and high risk. As a result, new technology has been developed that shows promise in alleviating these issues. Better blood tests, based on fragments of DNA suspended in blood, are set to enter the market in coming years. Early results indicate these tests will strongly outperform PSA and other protein based blood tests.
With these new tests on the horizon it is worth reflecting on the role imaging has in diagnostics and the outlook for its future.
Beginning way back in the early 19th century, blood based testing has long been a critical component in medical diagnosis. Some of the earliest tests included blood type testing, the advent of which enabled large scale transfusion efforts and has saved countless lives. As pathological science continued to develop, more and more conditions were tested using blood-based biomarkers. Today almost anyone can visit a GP, have bloods drawn and send them for analysis with little effort and cost. Many diagnostic tests of today including PSA, CA-125 and HER2 measure the presence of proteins in the blood, which can often indicate cancer or other diseases.
The downside with these tests though has been that many of them have correlates which are not related to disease. PSA for example is known to be higher in regular cyclists and can lead to false alarms. As a result, these tests are insufficiently accurate for large scale diagnostic screening. In the last 20 or so years genomic testing has come a long way. However, recently a number of companies have started commercialising tests based on the presence of blood based DNA. One example is ctDNA, which is DNA from tumor cells that can enter the bloodstream in the presence of a tumor. As tumors differ in genetics the sequencing of these ctDNA snippets can indicate and differentiate between tumors in the body. While not all of these tests are currently on the market, they certainly show great promise. A link in the latest Fortune Health newsletter points to the AACR’s opening plenary in which ctDNA and other blood based genomics received significant praise.
Why not Just Bloods?
With all these futuristic blood-based tests on the horizon, it would be easy to think diagnosis is almost a solved problem. Well, perhaps not easy but tempting. While these tests promise to provide a method of splitting the population into ‘sick’ and ‘well’, diagnostics focuses on knowing more than just whether someone is unwell. First and foremost blood based tests are currently unable to determine an exact location of any given tumor. Knowing whether it is in the brain or the prostate is a start but, if surgery or radiation therapy is the best treatment, then more needs to be done to plan for treatment. Today, medical imaging such as CT or MRI provides the best way of finding the exact location of tumors. In prostate cancer care, the number of patients who are now getting an MRI scan between a PSA test and a biopsy is on the rise. In other cancer such as lung or breast, imaging is a large and critical part of both screening and localisation.
Imaging is not without costs, both literally and otherwise. Imaging modalities utilizing X-Rays can, with long term exposure, cause health issues. This problem is well known and well managed but stops the technology from being used regularly. MRI scans have no known damaging radiation but typically come at a higher cost. Governments and private insurance companies are however, providing more in the way of rebates for MRI scans.
I spent my PhD designing MRI systems so naturally, they hold a special place in my heart, and while this isn’t the right post to talk about their inner workings, the fact that we can see inside a human body by manipulating magnetic fields is still quite astounding to me. MRI is becoming a bigger part of clinical practice worldwide and in developed countries are readily available in most places. Image quality is high and sequences can be adjusted to view not only anatomical but functional information about the patient. In the world of diagnostics, this means we can find tumors, make some inference about their functional state and plan treatment accordingly. This allows grading of tumors, separating those that need to be treated and those that are essentially harmless and pose little risk.
Currently, imaging is a key part of diagnosis that works together with blood-based tests to improve patient outcomes. Reading these images however, is a time consuming and complex task which still maintains a high false positive rate. Additionally, individual variability can be high meaning results from a single patient depend on the reader chosen. In this way computer aided diagnosis can help boost performance and ensure results are repeatable and accurate.
Imaging + ML
Machine Learning has seen large increases in performance over the last 5 years. Applications in image processing, speech processing and time series analysis have all seen new applications in medicine. Excitingly, recent applications have begun to outperform humans in image-based diagnosis, allowing fast reliable detection of skin cancer and diabetic retinopathy to name two of the larger works. With these and other projects, there is a growing opportunity for machine learning to boost the performance of today’s top clinicians. Many diseases in which imaging is a key component will no doubt see more machine learning move into clinical practice. While there is, as always, some push back from the clinical community on new technology, the promise of AI to boost speed and outcomes is bringing the clinical community around to the promise of AI.
The promise of AI in medicine is the driving force behind our work at Maxwell MRI. We have focused on prostate cancer as it affects almost 1 in 7 men and although we have many effective treatments for this disease, the shortfalls in diagnostic accuracy mean that many avoid diagnosis or receive diagnosis later than they should. Maxwell uses MRI scans and electronic health record data to assist in prostate cancer diagnosis and training our AI models.
Our initial results show above human performance and we are currently looking for US clinical partners to provide validation data to confirm these results.
Taking all this into consideration, we as humans are likely to see a fundamental shift in the way health care is delivered over the coming years, something I am excited to be a part of.
The future is bright. New blood tests are set to help us screen the entire population for things such as cancer. Non-invasive and non-radiation based methods let us confirm, localise and grade these results. All of that tied together with AI working alongside clinicians enables delivery of top tier health care to everyone.
That is the future I want to build, and the one I want future generations to share.