Visualizing Dementia: How Developments in CT and MR Imaging help differentiate dementia subtypes

 Current predictions estimate that dementia is slowly developing – invisibly –- within the brains of 35.6 million people worldwide1. That number is growing yearly, but despite its high prevalence and impact on patients’ lives, dementia remains vastly underdiagnosed. Due to the wide range of related yet clinically different syndromes, developing sufficiently specific and sensitive clinical criteria to identify and discriminate them is proving increasingly difficult. Neuroimaging is proving to be central in this pursuit. With recent advances in CT and MRI, combined with quantitative analysis using artificial intelligence (AI), better insight into dementia may only be a scan away.

The origins of neuroimaging

In the early 1900s, the first attempts at neuroimaging used a technique called pneumoencephalography. This technique involved withdrawing cerebrospinal fluid (CSF) via lumbar puncture and then replacing it with a gas (usually air) to outline the cortical surface and cerebral ventricles using X-ray roentgenography2. In the 1970s and 1980s, modern imaging techniques were developed, ushering in an age of more accessible, safer, non-invasive, painless, and (to a reasonable extent) repeatable neuroimaging2. These techniques include computed tomography (CT) and magnetic resonance imaging (MRI), which over the last decades have made significant advances in increasing spatial resolution, reducing noise and artefacts, and improving low-signal performance in scans3. These advances, combined with AI protocols, promise to bring a new revolution to neuroimaging.

CT and its role in dementia diagnosis

With the advent of computerized axial tomography (CAT or CT scans), highly detailed three-dimensional anatomical images of the brain are becoming readily available for diagnostic and research purposes. Consequently, the number of CT scans performed globally has also increased, with 3 million performed in the 1980s to 62 million performed in 20074. A proportion of this number can be explained by guidelines by the UK National Institute of Health and Care Excellence (NICE) and European and US diagnostic guidelines which now recommend evaluation of patients with suspected dementia to include structural imaging using non-contrast-enhanced CT or MRI5–7. The use of neuroimaging has thus shifted from solely the exclusion of potential secondary pathologies (such as cerebral tumors) to also investigate neuroanatomical features, which could help support a clinical diagnosis of the underlying cause(s) of dementia8. While currently there are no neuroanatomical features which provide perfect specificity and sensitivity for any given dementia syndrome, several structural patterns or signatures may provide positive predictive value and help physicians and clinicians narrow the differential diagnoses to most likely underlying pathologies9.

As a first-level examination of dementia, CT imaging is now often required6,10–12. While the technique is frequently regarded as “old fashioned”, it remains a valuable tool in detecting potential secondary, sometimes treatable, causes of cognitive impairment13. Typical examples include intracranial neoplasms, subacute and chronic subdural hematoma14, or normal pressure hydrocephalus15, to name a few. However, while CT is less sensitive than MRI in detecting cognitive impairment related changes, its value in some circumstances or centers revolve around its greater availability, lower cost, shorter acquisition time, making it adaptable to patients with poor compliance, and its utility for patients with metal devices, such as a pacemaker13.

Recent advancements in CT for dementia diagnosis

Modern developments in CT technology have aimed at optimizing image quality while reducing radiation dose3. This delicate balance is benefitting greatly from advancements in system properties and accompanying data acquisition and reconstruction techniques. System property improvements include smaller detector elements, resulting in improved spatial resolution, while multiple x-ray sources provide faster imaging and improved temporal resolution. While these improvements have demonstrated great promise in coronary CT angiography16, their application in assessing vascular changes in dementia remains an area of high interest17. Acquisition parameters further improved through dual-energy imaging, enabling numerous imaging types (weight average images, material decomposition images, electron density maps, etc.) in a single scan and new scanning procedures such as spiral scanning, which reduced the scan time. These new imaging capabilities have improved the detection of underlying (vascular and non-vascular) pathologies18. Ultimately, improved reconstruction parameters aim to influence and enhance image quality directly. The extended field of view increases the image matrix size and allows further visualization of surrounding skin and tissues. These parameters, combined with image enhancement tools aimed at reducing noise and artefacts, improve spatial resolution3.

Recent advances in CT imaging for dementia diagnosis

Most advances of the past few years have been within the scope of PET/CT imaging. PET imaging is exquisitely suited for dementia imaging, as Aβ protein plaques, generally thought to play a major role in the biological process of AD, are relatively easy to measure using radiotracers. Additionally, tau, another protein that has been shown to be associated with cognitive impairment, can be visualized on PET scans using different tracers. Hence, further development and improvement of tracers has been the aim of promising research during recent years showing the ability to provide pathologic staging19. 

The rapid pace of Magnetic Resonance Imaging (MRI) developments

In the late 1970s, the first head and body images were obtained using MRI. Since then it has become a widely used clinical imaging modality providing neuroradiologists and clinicians with tomographic images of excellent spatial resolution and tissue contrast. Recent advancements in the hardware and software of MRI are now focused on solving the current and future challenges of MRI; firstly, by making MRI more broadly available and applicable through faster acquisition, optimized workflows, and higher patient throughputs. Secondly, by identifying the needs of an ageing and multimorbid patient population through adaptation and tailoring of protocols and procedures; and lastly, by making MRI an integral part of modern precision medicine through the increased use of quantifiable data, defining image biomarkers and integrating AI algorithms20.

An example of applying MRI in modern precision medicine by deducting quantifiable data from the MRI scans and leveraging AI algorithms to compare patient results to population data: brain MRIs provide information on the atrophy rate of a patient, in this example images are representative of Alzheimer’s disease (modified from reference 21).

MRI and its role in dementia diagnosis

Diagnosis of dementia remains largely clinically supported by neuropsychological testing and aided by biomarker status determination through neuroimaging and/or CSF testing. The diagnostic certainty gained through these methods aim to provide improved prognostication and targeted symptomatic medication21. Importantly, while there is often overlapping clinical and imaging features in neurodegenerative disorders, characteristic patterns or “signatures” can be used to improve differential diagnosis22.

Increasingly vital in dementia (differential) diagnoses are structural MRIs (sMRIs). When a patient presents with suspected dementia, sMRIs aid in excluding secondary underlying pathologies, assessing brain atrophy and vascular burden, and identifying any markers or subtle changes in brain architecture to help differentiate between dementia syndromes. To this end, different imaging sequences are often used, such as T1*-weighted imaging to show characteristic patterns and signatures such as cortical atrophy, and their progression over time. Further, T2*-imaging is highly sensitive in detecting and locating microhemorrhages, bringing diagnostic clarity to underlying pathologies in different brain regions. For example, microhemorrhages in the cortical region are often Aβ-related while microhemorrhages associated with hypertension are found in deep brain regions23. Further subtle changes, such as metabolic, toxic, or infective processes, can also be visualized using T2*-weighted or FLAIR images and ultimately correlated with related cognitive deficits8. The further capabilities provided by MRI to identify and differentiate dementia subtypes through hippocampal and ventricular volumetric and shape analysis22 is greatly expanding our knowledge of the disease mechanisms.

The next step: using AI in visualizing dementia

Given the many common features across the dementias, subtle differences in syndromes remain difficult to distinguish in CT and MRI. Thus, a diagnosis often relies on using semi-quantitative scales while relying heavily on the experience of the neuroradiologist. A correct diagnosis, however, is becoming increasingly important in an era when potential disease-modifying agents are soon to be marketed24,25. To improve the diagnostic accuracy, AI software may play an increasingly crucial role in the diagnostic process, providing quantitative measurements to support diagnosis. AI Software, such as Quantib® Neurodegenerative, offers quantified segmentation of brain structures and white matter hyperintensities (WMH). This automated process allows for tracking and comparative analysis within quantified brain volumes, atrophy assessment, and WMH volume and number. These results can then give an automated determination of the semi-quantitative scales and consequently provide a differential diagnosis within a percentage of certainty. As more data is collected, these AI systems and algorithms improve, increasing their ability to accurately discriminate dementia-subtypes. In our blog, you can also read more about advancing Alzheimer’s MRIs using AI-based quantitative imaging.

Conclusion

Developments in (PET/)CT and MRI hold great potential for visualizing and identifying dementia-specific signatures, possibly even before the onset of cognitive decline. While dementia is a heterogeneous disease, the improved spatial and temporal resolution of neuroimaging is on the road to identifying subtle structural and vascular pathologies, enabling better diagnostic and biomarker-based discrimination between dementia syndromes. The additional potential of combing neuroimaging with AI-powered quantitative assessment may uncover currently unknown biomarkers and provide novel insight into our understanding of dementia.

How Aducanumab will impact neurology and neuroradiology practice

 Last week the FDA approval of the Alzheimer’s Disease (AD) focused drug aducanumab (or Aduhelm, the brand name chosen by Biogen) was all over the news. It is clear that the decision impacts many greatly, from AD patients, to relatives, to pharmaceutical companies and physicians alike. But what makes this approval so special? What are the claims and the doubts associated with aducanumab? And how does this influence the AD diagnosis and treatment trajectory from a neurology and neuroradiology standpoint?

 

What makes (the approval of) aducanumab so special?

Aducanumab is the first approved drug that has proven to clean-up amyloid-β plaques in the brain as seen on amyloid PET scans. These brain scans of patients receiving aducanumab showed a significant decrease in amyloid over the course of the treatment. Amyloid-β proteins are believed to be a root-cause of AD, although this hypothesis is not undisputed. Research has shown that elevated levels of the protein correlate with early dementia stages and cognitive decline or an increased risk thereof1–3. Hence, pursuing a drug that tackles amyloid-β (while research aimed at confirming the amyloid hypothesis further) seems a promising path to embark on. Which is exactly why aducanumab is not the only drug of this type in the pipeline. Many other pharmaceutical companies are working on similar therapies and it is not unlikely these will also seek FDA approval in the foreseeable future 4,5.

What are the claims made about aducanumab?

Two clinical trials underpin most of the evidence used in the FDA’s decision to approve aducanumab: the EMERGE trial and the ENGAGE trial. Both trials were halted early by Biogen because the effect of aducanumab on the endpoint CDR-SB (a neuro-psychiatric rating) was insignificant. However, after re-analyzing the results, a significant effect on the amyloid-β levels in the brain of the participants that received the highest drug dose was found6,7.

What causes doubts around aducanumab?

The end-point results were meager, causing an advisory panel to the FDA to vote against approval on November 6, 2020. Despite this advise the FDA decided to approve the drug through the accelerated approval process. This is a special procedure that can be applied when a drug targets a serious or life-threatening disease for which there is currently no sufficient therapy available. Or put differently: it is to be expected that the benefits of providing this drug to patients will outweigh the current downsides of not providing it. This means that the FDA requires an extensive phase 4 trial to confirm the hypotheses that cleaning-up amyloid-β results in better cognitive outcomes.8.

Another issue raising doubts are the side effects of aducanumab. Of all side effects ARIA (amyloid related imaging abnormalities, referring to cerebral edema or microhemorrhages) was by far the most common one, occurring in 41% of the subjects that received a high dose. However, of this 41% only 20% was symptomatic and Biogen claims that “the majority of patients who experienced ARIA were able to continue investigational treatment”. Nonetheless, symptoms of ARIA (headache, dizziness, visual disturbances, nausea and vomiting) can influence someone’s live severely6,9.

So what effect can neurologists expect from the introduction of aducanumab?

Now that aducanumab is clinically available in the United States, neurologists there can subscribe it and patients will most likely start asking for it. How will this impact the way neurologists diagnose and treat patients suffering from AD?

First of all many patients will likely ask for a more definitive (differential) diagnosis because they want to know whether they would be a candidate for the new treatment. Secondly patients will likely want to get the diagnosis earlier as it is hypothesized that the effect of aducanumab will likely be higher in a mild cognitive impairment (MCI) stage of the disease avoiding major damage from exposure to the amyloid-β plaque. Neurologists will have to be prepared for an increase in patients suspected of early stages of AD.

Third, patients are not blind the controversy surrounding the new treatment so neurologist will need to be prepared to clearly articulate the current status of this new treatment and the challenges involved with the approval.

Finally, the aducanumab treatment process is far from a one-time event. Patients will get the drug administered intravenously every four weeks and especially in the early phase of the therapy patients need to be monitored closely, resulting in more AD-related work for neurologists10,11.

And what will be the impact of aducanumab’s clinical introduction for neuroradiologists?

The AD diagnosis and treatment selection process will include brain scans, with the amyloid PET scan as the most important modality for determining the presence and amount of amyloid-β plaques. This scan is costly and invasive therefore is can be expected that lower-cost, less invasive MRI scans will start playing a bigger role in early detection of abnormal brain degeneration. Additionally, therapy monitoring will also require regular examination leveraging medical imaging. Current guidelines advise an MRI before the 7th and the 12th infusion11. To get the most information out of the MRI scans and to follow the disease development in detail, volumetry based on MRI is expected to gain importance. In conclusion, neuroradiologists will play an increasingly important role in diagnosis and treatment follow up of AD potentially in a similar fashion in how MS treatment options increased the workload when they became available9.        

What can AI do to support assessment of patients receiving aducanumab?

To manage the increased workload, it is important to consider AI support options early on. What solutions could be developed or are already available that will help physicians to handle the upcoming rise in Alzheimer’s related examinations?

Artificial intelligence has made its way to the healthcare sector already and will increasingly be implemented in medical facilities in the coming years. Also in the context of neurology and radiology various applications have seen the light of day and are increasingly adopted in clinical practice. To improve the care pathway for patients with an AD suspicion and ease the burden on involved clinicians AI could be leveraged in many ways, such as automated analysis of amyloid PET calculating relevant metrics for disease monitoring, but also algorithms that quantify results from MRI such as atrophy tracking or microbleed detection12.

Concluding

The approval of aducanumab is a big step for the medical field. It is the first amyloid-β targeted drug to make it to the market, and although the evidence used by the FDA is not extensive, biomarker results look promising and further trials will be implemented to gather more proof. It is safe to say that the availability of aducanumab, and most likely similar drugs to come, will influence diagnosis trajectories and treatment monitoring. AD-related workload will increase for both neurologists and neuroradiologists making it essential to keep innovating in this space to take away unnecessary burdens from the clinicians. At Quantib we believe AI will play a major role in quantifying disease onset as well as treatment effectiveness.

Bibliography

1. Näslund, J. Correlation Between Elevated Levels of Amyloid β-Peptide in the Brain and Cognitive Decline. JAMA 283, (2000).
2. Roe, C. M. et al. Amyloid imaging and CSF biomarkers in predicting cognitive impairment up to 7.5 years later. Neurology 80, (2013).
3. Rowe, C. C. et al. Predicting Alzheimer disease with β‐amyloid imaging: Results from the Australian imaging, biomarkers, and lifestyle study of ageing. Annals of Neurology 74, (2013).
4. Haeberlein, S. B. What have we learned from aducanumab? http://www.alz.org/documents_custom/2017-facts-and-figures.pdf. (2018).
5. Cummings, J. et al. Aducanumab produced a clinically meaningful benefit in association with amyloid lowering. Alzheimer’s Research & Therapy 13, (2021).
6. Haeberlein, S. B. et al. EMERGE and ENGAGE Topline Results: Two Phase 3 Studies to Evaluate Aducanumab in Patients With Early Alzheimer’s Disease. https://clinicaltrials.gov/ct2/show/NCT02477800.
7. Haeberlein, S. B. et al. Emerge and Engage topline results: Phase 3 studies of aducanumab in early Alzheimer’s disease. Alzheimer’s & Dementia 16, (2020).
8. Cavazzoni, P. FDA’s Decision to Approve New Treatment for Alzheimer’s Disease. https://www.fda.gov/drugs/news-events-human-drugs/fdas-decision-approve-new-treatment-alzheimers-disease (2021).
9. Scheltens, P. Extra Lunch & Learn over toelating alzheimermedicijn. (2021).
10. Scheltens, P. Op1. (2021).
11. Medscape. Aducanumab. https://reference.medscape.com/drug/aduhelm-aducanumab-4000138.
12. Mormino, E. C. & Papp, K. v. Amyloid Accumulation and Cognitive Decline in Clinically Normal Older Individuals: Implications for Aging and Early Alzheimer’s Disease. Journal of Alzheimer’s Disease 64, (2018).

Can AI solve radiologist’s most common challenges in prostate MRI?

 In the past years, different research projects have been conducted in an effort to find a more efficient and accurate way of carrying out the clinical pathway for prostate cancer (PCa) diagnosis.

Thanks to the results of the PROMIS study, reported by The Lancet in 20171, which demonstrated how beneficial including a prostate MRI can be in the diagnostic process of PCa, organizations such as the EAU (European Association of Urology)2 and the ACR (American College of Radiology)3 changed their guidelines for PCa diagnosis to include the acquisition of an MRI pre-biopsy as a standard procedure.

Naturally, the number of prostate MRIs expected to be performed every year will exponentially increase. Which brings the radiology community to a crossroads: is it possible to maintain the quality of patient care taking into account this increase of workload?

The answer is yes, with the right strategy. The following blog will detail what challenges occur together with the increase in workload and how an Artificial Intelligence (AI) software solution for reading prostate MRI can help radiologists overcome the most frequent pain points in the field.

Prostate MRI and the radiologist’s challenges

One of the biggest struggles for professionals in the prostate field is how time-consuming the whole process is, from logging in to all required software for the reading process, to the assessment itself, dictating and report creation. Guidelines are available, however, the standardized process includes a long list of steps, which can be tedious and demanding.

Additionally, many software programs that are available for MRI scan assessment do not offer all functionalities needed for precise evaluation of prostate MRIs. For example, the user would have to use a PACS viewer to look at the images, use an online calculation tool to determine the PSA density and use yet another software to dictate their report. This makes the process even more laborious as radiologists have to change back and forth between different programs.

For example, many facilities require a measurement of prostate volume to calculate prostate density. However, the traditional measurement method, using a formula that approaches the shape of the prostate gland with an ellipsoid, usually requires the radiologist to measure the prostate gland in multiple dimensions in their reading software, after which they use an online tool to estimate the prostate volume. In addition, this approach has been proven to provide relatively inexact results4,5. In fact, research appoints the ellip

soid formula to be the primary reason for inconsistency in prostate volume estimations resulting, mainly, in underestimations of the gland size.5

Moreover, these programs often lack the user-friendliness that would make the work of a radiologist much more enjoyable. Often software tends to be slow, non-intuitive and in some cases it can even suggest imprecise results. For example, the speech software connected to the Dictaphone often provides erroneous texts requiring the radiologist to correct some parts of the final report manually.

Lastly, it is important to note the difficulty that reading prostate MRI presents for less experienced radiologists. Correctly assessing these MRI scans of the prostate requires a level of expertise that comes from years of experience6. Hence, starting radiologists will need to assess many exams before reporting on MRIs with a certain level of ease and confidence.Without AI a radiologist starts his learning curve when he/she starts medical school (blue solid line). The gained expertise for reading prostate MRIs will increase in time, with the strongest slope during sub-specialty training. Implementing AI software to support radiologists will allow them to use highly specialized knowledge earlier on in their careers.

What can AI do to help the radiologist?

There are many ways in which AI can support radiologists with the challenging task of reading prostate MRIs. Providing an all-round software package offering a range of assisting AI functionalities will make the lives of radiologists easier and, most likely, more efficient, as performing daily tasks with well-designed and comprehensive solutions can provide an efficiency boost to day-to-day task.

Firstly, AI software can avoid the radiologists’ need to do their assessment divided over many places. For example, a software, thank to their AI algorithms, can automatically, and in seconds, determine prostate volume and PSA density, which not only eliminates the imprecise ellipsoid approach from the process, but also saves the radiologist time.

Moreover, the software can also provide automated analysis that assesses multiple sequences in one go. Such information can be presented in an intuitive heatmap making it easier to visualize the combined information and highlighting possible regions of interest.

Furthermore, AI software can streamline the workflow, for example, by providing step-wise guidance for the determination of the PI-RADS score. Each essential input will be requested by the software, guiding the user through the PI-RADS guidelines and providing valuable support to get to a final PI-RADS scoring.

With all support AI software can provide, such a program can be specifically useful for starting radiologists to improve their reading skill early on. Nonetheless, prostate MRI can be challenging field for radiologists of all levels of expertise, especially if we take into account the increase in workload previously mentioned. Stress and mental fatigue can lead to inter- and intra- observer variability which can negatively impact the efficiency of the radiologist workflow. However, functioning as a second opinion, the automated calculations and analysis AI provides can reduce that variability.

The automated generation of a standardized report can be a significant asset for multidisciplinary team meetings (MDTs). The report can include key images, which otherwise may take additional time to collect and save. In case of some PACS vendors, this time may even have to be spent during MDT, as images need to be looked up on the spot. Additionally, easily understandable graphs and metrics provide valuable context for decision making. Consequently, the amount of time radiologists need to prepare for these meeting will be reduced and the meetings themselves will be more time-efficient.

How can AI minimize the urologist’s challenges associated with prostate MRIs?

AI solutions for prostate MRI reading may result in a more efficient information exchange between radiologists and urologists.

As an AI software can reduce the amount of time it takes radiologists to assess the images and report their findings, the turn around time for the radiology exam may be shortened as well as results which translates into less waiting time for the patient to receive the outcome of their diagnosis trajectory.

Moreover, the decrease in inter- and intra- observer variability will make the results easier to interpret by a urologist and easier for him to make better informed decisions about the next steps in the diagnostic process of their patients. During communication with the patient, the AI-generated report may also provide an easy-to-understand base to support the urologist explaining the situation and treatment option.

Conclusion

These many challenges, from the heavier workload to the non-intuitive interfaces, present a significant risk to the mental health of the professionals involved in the clinical pathway for PCa diagnosis. As these challenges can lead to continuous stress and burnout, it is essential to tackle these challenges to avoid a decrease in quality of patient care.

Although recent research7 shows there is always room for improvement when it comes to the application of AI to prostate MRI, incorporating AI into the radiology workflow may solve many challenges starting today. Therefore, AI driven solutions for radiology can play a key role in improving the day-to-day conditions these professionals encounter and advance the quality of patient care.

6 use cases for CT body composition with AI

 The physical state of a patient is relevant to many different areas of healthcare. A fast, commonly used metric for this is the body mass index (BMI, weight divided by height in meters squared). However, BMI only provides an indication of how the weight of a patient relates to their height and does not give any information about muscle mass or fat distribution. Therefore, it is a poor estimate of a patient's physical state.

A better and more precise measure of a person's physical state is his or her body composition: the body's layout in terms of muscles, fat and organs. This much more informative measure can be obtained using information in medical imaging studies acquired for other purposes. For example CT scans are routinely acquired tens of thousands of times per year in any hospital. However, deriving the body compositional information from these scans would require manually delineating the tissues depicted in the scan; a very time-consuming task. While researchers have long known that this method could work, it is not feasible to perform in clinical practice. Fully automatic segmentation of these CT scans could therefore be of great help in providing accurate measurements of body composition based on imaging data that is readily available.

Figure 1: Body composition analysis can offer valuable additional value to BMI calculation alone.

A large range of different use cases are relevant in the context of body composition analysis. Below are six examples where such automatic measurements could be of significant clinical value, including cardiovascular risk analysis, oncology and COVID-19 prognostics.

AI body composition for cardiovascular risk analysis

Many studies found that body composition is an important risk factor in cardiovascular disease.1 Automatic body composition analysis could therefore provide additional biomarkers for cardiovascular risk assessment.

People with a similar BMI can have a very different cardiometabolic risk profile. Especially people with excess visceral fat are at a high risk, which is not captured by solely measuring their BMI. Increased amounts of visceral fat are associated with e.g. type 2 diabetes, atherosclerosis and cardiovascular disease. Furthermore, excess visceral fat is often also associated with micro- or macroscopic fat inside of other tissues (ectopic fat), such as in or around skeletal muscles, in the liver and around the heart, which poses additional health risks – and is predictive of outcome in multiple clinical settings.1. CT is an excellent modality to allow accurate measurements of fat distribution in the body and therefore a highly suited modality for body compositional measurements which can be easily obtained using automated AI-powered tools. 

AI body composition for oncology

A large number of studies have shown the predictive power of body composition on outcome in cancer patients.2 Accurate analysis of body composition has been shown to improve prognostication and early identification of patients at risk: for example, decreased muscle mass (sarcopenia) predicts adverse outcomes in gastric cancer, hepatocellular carcinoma and metastatic renal cell carcinoma3.

Moreover, treatment response to chemotherapy or immunotherapy is highly influenced by body composition.4,5 Skeletal muscle depletion has been recognized as an independent predictor of toxicity. Accurate AI-based body composition measurements could therefore also help in better estimating the required dose for each individual patient. Such personalized dose estimation could then prevent toxicity, improve efficacy and reduce costs of cancer treatments.

AI body composition for surgery

Another promising application of automatic body composition analysis is in preoperative risk assessment. For example in cardiovascular surgery, the predictive nature of body compositional features has been shown in patients undergoing transcatheter aortic valve implantation, where psoas muscle area analyzed on preoperative CT scans has been found to directly predict postoperative mortality.6 In surgical oncology, sarcopenia, skeletal muscle fat infiltration and visceral obesity measures derived from body composition analysis, predicted postoperative complications in patients undergoing gastrectomy for gastric cancer.7 Hence, deploying AI to support surgery preparations holds great potential for improved patient outcomes.

AI body composition for contrast dose estimation

Furthermore, body composition analysis could improve the estimation of required contrast dose in CT, MRI and PET scans. It has been shown that estimation of lean body mass can help to reduce iodine dose in CT, while maintaining the required level of attenuation.8 Commonly, lean body mass is estimated based only on total body weight and height. In multiphase CT scans, the required contrast dose could therefore be accurately estimated based on a fast AI-powered body composition analysis of the precontrast image.

AI body composition for chronic kidney disease

A related application is chronic kidney disease, where automatic body composition analysis can both help predict renal function, which derives from total muscle mass9, as well as accurately quantify muscle wasting, a frequent finding in kidney disease10. Muscle wasting is associated with physical disability, worse quality of life and increased morbidity. Accurate analysis of body composition can help in making personalized treatment decisions for these patients. Hence, automating and therefore standardizing these analyses using AI may advance treatment trajectories considerably.

AI body composition for COVID-19

Finally, in light of the recent COVID-19 pandemic, body composition has also been recognized as a prognostic factor for patients with COVID-1911. In this study, access to CT body composition features allowed doctors to quantify the ratio between long spine muscle circumference and waist circumference. This measurement was found to predict which patients would end up in the intensive care unit, making CT body composition a cost-effective method to estimate required clinical resources, especially when analysis can be automated using AI.

In conclusion

Body composition measurements are a very promising tool in a wide range of use cases – a tool that is now available to any physician or researcher through fully automatic AI-based analysis of any CT scan.

Interested to try out AI-powered body composition measurements on your own CT data? Get in touch!

 

1. Neeland, I. J. et al. Visceral and ectopic fat, atherosclerosis, and cardiometabolic disease: a position statement. Lancet Diabetes Endocrinol. 7, 715–725 (2019).
2. Brown, J. C., Cespedes Feliciano, E. M. & Caan, B. J. The evolution of body composition in oncology—epidemiology, clinical trials, and the future of patient care: facts and numbers. J. Cachexia. Sarcopenia Muscle 1200–1208 (2019) doi:10.1002/jcsm.12379.
3. Shachar, S. S., Williams, G. R., Muss, H. B. & Nishijima, T. F. Prognostic value of sarcopenia in adults with solid tumours: A meta-analysis and systematic review. Eur. J. Cancer 57, 58–67 (2016).
4. Prado, C. M. M. Body composition in chemotherapy: The promising role of CT scans. Curr. Opin. Clin. Nutr. Metab. Care 16, 525–533 (2013).
5. Daly, L. E. et al. The impact of body composition parameters on ipilimumab toxicity and survival in patients with metastatic melanoma. Br. J. Cancer 116, 310–317 (2017).
6. van Mourik, M. S. et al. CT determined psoas muscle area predicts mortality in women undergoing transcatheter aortic valve implantation. Catheter. Cardiovasc. Interv. 93, E248–E254 (2019).
7. Zhang, Y. et al. Computed tomography–quantified body composition predicts short-term outcomes after gastrectomy in gastric cancer. Curr. Oncol. 25, 411–422 (2018).
8. Matsumoto, Y. et al. Contrast Material Injection Protocol With the Dose Determined According to Lean Body Weight at Hepatic Dynamic Computed Tomography: Comparison Among Patients With Different Body Mass Indices. J. Comput. Assist. Tomogr. 43, 736–740 (2019).
9. Donadio, C. Body composition analysis allows the prediction of urinary creatinine excretion and of renal function in chronic kidney disease patients. Nutrients 9, (2017).
10. Sabatino, A. et al. Muscle mass assessment in renal disease: The role of imaging techniques. Quant. Imaging Med. Surg. 10, 1672–1686 (2020).
11. Kottlors, J. et al. Body composition on low dose chest CT is a significant predictor of poor clinical outcome in COVID-19 disease - A multicenter feasibility study. Eur. J. Radiol. 132, 109274 (2020).

AI in prostate cancer: 6 'must-read' papers

 Since recent guidelines advise the acquisition of an MRI prior to prostate biopsy, the number of prostate MRI scans made has further increased. Hence, timing could not be better for methods that support the radiologist with their analysis, for example, by (partly) automating the reading process. Artificial intelligence (AI) algorithms have been proven to successfully speed up complex tasks in other industries, which is why academic researchers are now actively investigating the applications of AI in prostate cancer diagnosis.

In this post we discuss 6 recent inspiring articles providing a flavor of what AI can do for diagnosing prostate cancer.

 

Let’s start with an overview of AI in prostate cancer

A Systematic Review of Artificial Intelligence in Prostate Cancer
Booven, D. J. Van et al.

For an overview of the already abundant possibilities, “A Systematic Review of Artificial Intelligence in Prostate Cancer” is a nice place to start. Van Booven et al. provide a compact overview of the many opportunities of AI in the diagnostic pathway of prostate cancer.

Starting with AI-based analysis of PSA measurements and other patient data to increase the sensitivity and specificity versus PSA determination alone. Additionally, applying artificial neural networks to histopathological data is discussed, for example, by predicting biopsy results and determining the Gleason score. This type of analysis offers the opportunity to reduce inter- and intra-observer variability versus human assessments. A third field of application that is addressed is the AI-assessment of prostate MRI scans to determine the Gleason scores.

To aid research of new (non-imaging) biomarkers, AI can play an important role in assessing a whole range of gene expressions that may be related to prostate cancer occurrence. As datasets can become extremely large, having AI lend a helping hand in this situation may be a godsend.

A less frequently discussed field, but nonetheless an area that holds great promise to improve prostate cancer care, is the scope of patient-centered treatment. For example by leveraging AI to compare patient data to norms derived from large databases. This will allow patients to better understand their situation and be in charge of their care. It may also bolster standardized data sharing, offering great opportunities to improve PCa care.

Lastly, van Booven et al. mention the great promise AI holds for stratification of risk of recurrence of prostate cancer. AI's ability to take many factors into account when determining a patient’s risk at cancer recurrence will aid in the search for new metrics.2

Applying AI to mpMRI for PCa diagnosis

Machine learning applications in prostate cancer magnetic resonance imaging
Cuocolo, R. et al.

Focusing specifically on AI applications in the context of prostate MRI Cuocolo et al. provide another excellent review in “Machine learning applications in prostate cancer magnetic resonance imaging”. The authors establish a basic understanding of machine learning by providing a very brief introduction (quite handy if you are unaware of the differences between supervised, unsupervised and semi-supervised learning, or are interested in a very brief primer of deep learning!). Additionally, the article discusses the typical pipelines for training a machine learning or deep learning algorithm in the context of prostate MRI. However, most ink is spent on an overview of the different applications of AI in prostate cancer, ranging from image segmentation to staging and biochemical recurrence. An extensive reference list supports the review and provides inspiration for further reading.3 Worth a read if you are interested in all the opportunities of AI in MRI prostate

A Deep Learning-Based Approach for the Detection and Localization of Prostate Cancer in T2 Magnetic Resonance Images
Alkadi, R. et al.

Alkadi et al. wrote a more technical paper on the segmentation of the prostate organ, its zones and tumors. They used a deep neural network with a “deep convolutional encoder-decoder architecture”, which basically comes down to the more popular Unet, however, to limit the necessary computational resources, authors used a more down-scaled version by cutting back on specific connections that are usually made between the very first layers (for down scaling of the image) of a Unet and the deeper layers (for upscaling of the image). Nice results are claimed when comparing tumor segmentations to a manually determined ground truth: AUC of 0.995, an accuracy of 0.894, and a recall of 0.928. While unique in its approach, it is nice for those interested in deep learning methods for segmentation of various part of the prostate.4

Deep-Learning-Based Artificial Intelligence for PI-RADS Classification to Assist Multiparametric Prostate MRI Interpretation: A Development Study
Sanford, T. et al.

One of the fundamental steps in the prostate MRI workflow is the determination of a PI-RADS score per lesion. “Deep-Learning-Based Artificial Intelligence for PI-RADS Classification to Assist Multiparametric Prostate MRI Interpretation: A Development Study” describes the training and testing of a deep learning network (a CNN, or for the techies amongst us, a ResNet) and its ability to automatically determine a PI-RADS score (for lesions that have already been identified and segmented on the MRI, which could also be additional tasks AI is able to tackle). Authors claim a similar performance for their algorithm to that of human experts, providing another interesting use case for AI application in prostate MRI.5

Prostate PET supported by AI

Artificial intelligence-based versus manual assessment of prostate cancer in the prostate gland: a method comparison study.
Mortensen, M. A. et al.

Next to applying AI to prostate MRI, PET of the prostate offers an attractive scope for AI deployment. An example paper by Mortensen et al. shows how a CNN can fully automate the segmentation of the prostate gland on CT images, where after the segmentation is aligned to the PET scan and elementary PET measurements are calculated. As the algorithm was able to provide perform these calculations in a matter of seconds with similar values to those manually determined, such applications offer a great promise of accelerating PET analyses.6

 Mortensen et al. trained a 2-step algorithm to, as a final output, provide segmentation of the prostate gland on both the CT and PET scan. Firstly the PET scan is registered to the CT scan, secondly the prostate gland is segmented on the CT scan and the registration transformation is used to determine the segmentation on the PET scan.

One step further: leverage AI to link MRI to pathology findings

Artificial intelligence at the intersection of pathology and radiology in prostate cancer
Harmon, S. A., Choyke, P. L. & Türkbey, B.

Last but not least the application of AI to pathology holds great promise. Inspecting biopsy samples using image analysis software offers the opportunity of accelerating the biopsy analysis process, as well as minimizing inter- and intra-observer variance.

In their review paper Harmon et al. discuss the application of AI to mpMRI scans and to pathology data, as well as the exciting option to apply AI to both mpMRI and pathological data in order to improve prediction and grading of prostate cancer cases. Briefly mentioned already under the summary of the first article in this blogpost, training an algorithm that can predict pathology results using mpMRI scans, will allow for a faster, non-invasive diagnosis trajectory. Harmon et al. shed some more light on this option, while stressing the fact that sufficient high quality data is needed to make such an algorithm to the trick with a sufficient level of performance. Read this review article to learn more about recent advances in AI applied to combining mpMRI with digital pathology and the possibility of determining advanced characterization using mpMRI alone.7

The radiology workflow for prostate cancer diagnosis consists of many steps. AI can offer a wide range of options to support this process.

Concluding

Apart from it being inspiring to read about all the possibilities and the great research that is done, the fact the results are now starting to make their way into the clinic is even exciting.

Why to implement AI for prostate MRI now?

 Most of us are familiar with the adoption curve. It is a nice arch resembling a normal distribution with on one end of the spectrum the innovators, there are only a few of those, then the early adopters, a slightly larger group, then a big bunch that belongs to the early and late majority, and closing off with the laggards.

This theory holds true in many industries, including healthcare. Over the past few years, we have seen an immense rise of artificial intelligence (AI) applied to healthcare cases, with radiology being a praiseworthy frontrunner. Many applications within this field have seen the light of day, ranging from image reconstruction to image analysis to support for standardized reporting. With so many AI software options, it may be hard to choose where to start with the implementation. It may be even tempting to wait and hold off for a while until other hospitals have tried it first, more scientific proof is gathered, or AI companies have spent much more development hours on optimizing their user experience.

In this article, we discuss why this is exactly what you should not do in the case of prostate MRI reporting. Continue reading if you are wondering how AI for prostate MRI can make radiologists’ lives easier and how it can do so starting tomorrow.

Become an expert in prostate MRI

Reading prostate MRIs is not an easy task. It involves a long learning curve, meaning it takes time to master the skill. AI software for prostate MRI reading is generally developed using a large batch of MRI scans, a number so large, that it will take years, if not more for a radiologist to process an equal amount of scans and generate the associated expertise. Training software to extract this knowledge, and for it to be able to apply it to new patient data and provide valuable insights will offer radiologists know-how and support during their reading process. Even a novice radiologist will have access to expertise which will otherwise need years to gain.

Without AI a radiologist starts his learning curve when he/she starts medical school (blue solid line). The gained expertise for reading prostate MRIs will increase time, with the strongest slope during sub-specialty training. Implementing AI software to support radiologists to use highly specialized knowledge earlier on in their careers.

Reduce workload while prostate MRI volumes increase

Speak to any abdominal radiologist and they will confirm: the amount of prostate MRIs acquired yearly is growing. Changes in the guidelines both in the US and in Europe advise for an MRI pre-biopsy in the diagnostic pathway of prostate cancer. Naturally, this increase in scans leads to an increase in workload, which will become a problem for radiologists to manage as the number of hours in a day did not change. Educating significantly more radiologists to perform the challenging task of reading prostate MRIs would be one way to handle this issue, however, speeding up the scan assessment and report creation process to decrease the time spent per case is most likely a much more feasible option. AI offers opportunities to optimize many steps in the reading process. What to think of automated PSA density calculation which is even more accurate as they are based on a precise segmentation instead of an ellipsoid-based calculation that only approximates the volume of the prostate gland? Or smart heatmaps that help radiologists quickly scan prostate MRIs looking for abnormalities? Another useful assisting tool is PI-RADS scoring support, providing easy step-wise implementation of the assigning of a PI-RADS score. As a last example: automated report creation, where the software generates an exhaustive, user-friendly report making the urologist smile from ear to ear.

Shout it from the rooftops: we use state-of-the-art AI for prostate MRI!

Using advanced technology for diagnosis and treatment draws both physicians and patients. Look at the famous surgical robot: surgeons are eager to work with it and patients become more and more eager to have their surgery performed using this piece of top notch tech. AI for image analysis has not been widely used to advertise (yet), while it can work wonders to draw more talent to apply at a hospital. Additionally, a medical institute that broadcasts that it standardly deploys AI software to support the medical image analyses process and provide a built-in second opinion, may inspire more patients to choose this facility for their care. After all, which patient would not like to have a standard second opinion built into the diagnostic process?

Design the future of AI for prostate MRI

AI radiology companies are constantly improving their software. This means there is a constant hunt for valuable input on how best to advance these products. Input best given by radiologist, on desired software features, which button should go where, preferred integration options, etc. Hence getting involved early in the process will provide radiologists, urologists, and other physicians involved with a chance to influence the way the software will look and operate in the future. In other words, it provides radiologists with an opportunity to shape their own future.

An additional advantage is that there may still be room for negotiation on price. The more extensive the software gets with development over time, the higher the development costs, the more expensive the price tag. Jumping on board now may help you secure a better price.

Don’t risk staying behind on AI for prostate MRI

AI in healthcare has shown impressive growth over the past few years. It is safe to assume that it will be implemented, maybe not tomorrow, but for sure in the coming years and decades, it will increasingly be part of patient care in one way or another, until it can no longer be ignored. One can decide to procrastinate and push the adoption of AI at their healthcare facility forward. But why postpone until tomorrow what can benefit you today?