Two studies show the accuracy of perception-based systems in detecting disease biomarkers without needing molecular recognition elements, such as antibodies
Researchers from multiple academic and research institutions have collaborated to develop a non-conventional machine learning-based technology for identifying and measuring biomarkers to detect ovarian cancer without the need for molecular identification elements, such as antibodies.
Traditional clinical laboratory methods for detecting biomarkers of specific diseases require a “molecular recognition molecule,” such as an antibody, to match with each disease’s biomarker. However, according to a Lehigh University news release, for ovarian cancer “there’s not a single biomarker—or analyte—that indicates the presence of cancer.
“When multiple analytes need to be measured in a given sample, which can increase the accuracy of a test, more antibodies are required, which increases the cost of the test and the turnaround time,” the news release noted.
Unveiled in two sequential studies, the new method for detecting ovarian cancer uses machine learning to examine spectral signatures of carbon nanotubes to detect and recognize the disease biomarkers in a very non-conventional fashion.
Perception-based Nanosensor Array for Detecting Disease
In the Science Advances paper, the researchers described their development of “a perception-based platform based on an optical nanosensor array that leverages machine learning algorithms to detect multiple protein biomarkers in biofluids.
“Perception-based machine learning (ML) platforms, modeled after the complex olfactory system, can isolate individual signals through an array of relatively nonspecific receptors. Each receptor captures certain features, and the overall ensemble response is analyzed by the neural network in our brain, resulting in perception,” the researchers wrote.
“This work demonstrates the potential of perception-based systems for the development of multiplexed sensors of disease biomarkers without the need for specific molecular recognition elements,” the researchers concluded.
In the Nature Biomedical Engineering paper, the researchers described a fined-tuned toolset that could accurately differentiate ovarian cancer biomarkers from biomarkers in individuals who are cancer-free.
“Here we show that a ‘disease fingerprint’—acquired via machine learning from the spectra of near-infrared fluorescence emissions of an array of carbon nanotubes functionalized with quantum defects—detects high-grade serous ovarian carcinoma in serum samples from symptomatic individuals with 87% sensitivity at 98% specificity (compared with 84% sensitivity at 98% specificity for the current best [clinical laboratory] screening test, which uses measurements of cancer antigen 125 and transvaginal ultrasonography,” the researchers wrote.
“We demonstrated that a perception-based nanosensor platform could detect ovarian cancer biomarkers using machine learning,” said Yoona Yang, PhD, a postdoctoral research associate in Lehigh’s Department of Chemical and Biomolecular Engineering and co-first author of the Science Advances article, in the news release.
How Perception-based Machine Learning Platforms Work
According to Yang, perception-based sensing functions like the human brain.
“The system consists of a sensing array that captures a certain feature of the analytes in a specific way, and then the ensemble response from the array is analyzed by the computational perceptive model. It can detect various analytes at once, which makes it much more efficient,” Yang said.
“SWCNTs have unique optical properties and sensitivity that make them valuable as sensor materials. SWCNTS emit near-infrared photoluminescence with distinct narrow emission bands that are exquisitely sensitive to the local environment,” the researchers wrote in Science Advances.
“Carbon nanotubes have interesting electronic properties,” said Daniel Heller, PhD, Head of the Cancer Nanotechnology Laboratory at Memorial Sloan Kettering Cancer Center and Associate Professor in the Department of Pharmacology at Weill Cornell Medicine of Cornell University, in the Lehigh University news release.
“If you shoot light at them, they emit a different color of light, and that light’s color and intensity can change based on what’s sticking to the nanotube. We were able to harness the complexity of so many potential binding interactions by using a range of nanotubes with various wrappings. And that gave us a range of different sensors that could all detect slightly different things, and it turned out they responded differently to different proteins,” he added.
The researchers put their technology to practical test in the second study. The wanted to learn if it could differentiate symptomatic patients with high-grade ovarian cancer from cancer-free individuals.
The research team used 269 serum samples. This time, nanotubes were bound with a specific molecule providing “an extra signal in terms of data and richer data from every nanotube-DNA combination,” said Anand Jagota PhD, Professor, Bioengineering and Chemical and Biomolecular Engineering, Lehigh University, in the news release.
This year, 19,880 women will be diagnosed with ovarian cancer and 12,810 will die from the disease, according to American Cancer Society data. While more research and clinical trials are needed, the above studies are compelling and suggest the possibility that one day clinical laboratories may detect ovarian cancer faster and more accurately than with current methods.
The study was partially retrospective in that the
researchers compiled past alerts generated by the CDS systems at BWH and MGH
between 2009-2011 and added them to alerts generated during the active part of
the study, which took place from January 1, 2012 to December 31, 2013, for a
total of five years’ worth of CDS alerts.
They then sent the same patient-encounter data that generated those CDS alerts to a machine learning platform called MedAware, an AI-enabled software system developed in Ra’anana, Israel.
MedAware was created for the “identification and prevention
of prescription errors and adverse drug effects,” notes the study, which goes
on to state, “This system identifies medication issues based on machine
learning using a set of algorithms with different complexity levels, ranging
from statistical analysis to deep learning with neural networks. Different
algorithms are used for different types of medication errors. The data elements
used by the algorithms include demographics, encounters, lab test results,
vital signs, medications, diagnosis, and procedures.”
The researchers then compared the alerts produced by
MedAware to the existing CDS alerts from that 5-year period. The results were
According to the study:
“68.2% of the alerts generated were unique to
the MedAware system and not generated by the institutions’ CDS alerting system.
“Clinical outlier alerts were the type least
likely to be generated by the institutions’ CDS—99.2% of these alerts were
unique to the MedAware system.
“The largest overlap was with dosage alerts,
with only 10.6% unique to the MedAware system.
“68% of the time-dependent alerts were unique to
the MedAware system.”
Perhaps even more important was the results of the cost
analysis, which found:
“The average cost of an adverse event
potentially prevented by an alert was $60.67 (range: $5.95–$115.40).
“The average adverse event cost per type of
alert varied from $14.58 (range: $2.99–$26.18) for dosage outliers to $19.14
(range: $1.86–$36.41) for clinical outliers and $66.47 (range: $6.47–$126.47)
for time-dependent alerts.”
The researchers concluded that, “Potential savings of $60.67 per alert was mainly derived from the prevention of ADEs [adverse drug events]. The prevention of ADEs could result in savings of $60.63 per alert, representing 99.93% of the total potential savings. Potential savings related to averted calls between pharmacists and clinicians could save an average of $0.047 per alert, representing 0.08% of the total potential savings.
“Extrapolating the results of the analysis to the 747,985
BWH and MGH patients who had at least one outpatient encounter during the
two-year study period from 2012 to 2013, the alerts that would have been fired
over five years of their clinical care by the machine learning medication
errors identification system could have resulted in potential savings of
Savings of more than one million dollars plus the prevention
of potential patient harm or deaths caused by thousands of adverse drug events
is a strong argument for machine learning platforms in diagnostics and
prescription drug monitoring.
Researchers Say Current Clinical Decision Support Systems
Machine learning is not the same as artificial intelligence. ML is a “discipline of AI” which aims for “enhancing accuracy,” while AI’s objective is “increasing probability of success,” explained Tech Differences.
Healthcare needs the help. Prescription medication errors cause patient harm or deaths that cost more than $20 billion annually, states a Joint Commission news release.
CDS alerting systems are widely used to improve patient
safety and quality of care. However, the BWH-MGH researchers say the current
CDS systems “have a variety of limitations.” According to the study:
“One limitation is that current CDS systems are rule-based and can thus identify only the medication errors that have been previously identified and programmed into their alerting logic.
“Further, most have high alerting rates with many false positives, resulting in alert fatigue.”
Commenting on the value of adding machine learning
medication alerts software to existing CDS hospital systems, the BWH-MGH
researchers wrote, “This kind of approach can complement traditional rule-based
decision support, because it is likely to find additional errors that would not
be identified by usual rule-based approaches.”
However, they concluded, “The true value of such alerts is
highly contingent on whether and how clinicians respond to such alerts and
their potential to prevent actual patient harm.”
Future research based on real-time data is needed before machine
learning systems will be ready for use in clinical settings, HealthITAnalytics
However, medical laboratory leaders and pathologists will
want to keep an eye on developments in machine learning and artificial
intelligence that help physicians reduce medication errors and adverse drug
events. Implementation of AI-ML systems in healthcare will certainly affect
clinical laboratory workflows.
Machine learning software may help pathologists make earlier and more accurate diagnoses
In Boston, two major academic centers are teaming up to apply big data and machine learning to the problem of diagnosing cancers earlier and with more accuracy. It is research that might have major implications for the anatomic pathology profession.
A collaborative effort between teams at Beth Israel Deaconess Medical Center (BIDMC) and Harvard Medical School (HMS) has resulted in an innovation that could result in more accurate diagnoses in the pathology laboratory. The teams have been working on a machine learning software program that will eventually function as an artificial intelligence (AI) to improve the accuracy of diagnostics. They hope to someday build AI-powered computer systems that can accurately and quickly interpret pathology images. (more…)
Pathologists and clinical laboratory scientists may find one hospital’s use of a machine-learning platform to help improve utilization of lab tests both an opportunity and a threat
Variation in how individual physicians order, interpret, and act upon clinical laboratory test results is regularly shown by studies in peer-reviewed medical journals to be one reason why some patients get great outcomes and other patients get less-than-desirable outcomes. That is why many healthcare providers are initiating efforts to improve how physicians utilize clinical laboratory tests and other diagnostic procedures.
This effort came about after clinical and administrative leadership at Flagler Hospital realized that only about one-third of its physicians regularly followed certain medical decision-making guidelines or clinical order sets. Armed with these insights, staff members decided to find a solution that reduced or removed variability from their healthcare delivery.
Reducing Variability Improves Care, Lowers Cost
Variability in physician care has been linked to increased healthcare costs and lower quality outcomes, as studies published in JAMA and JAMA Internal Medicine indicate. Such results do not bode well for healthcare providers in today’s value-based reimbursement system, which rewards increased performance and lowered costs.
Clinical order sets are designed to be used as part of clinical decision support systems (CDSS) installed by hospitals for physicians to standardize care and support sound clinical decision making and patient safety.
However, when doctors don’t adhere to those pre-defined standards, the results can be disadvantageous, ranging from unnecessary services and tests being performed to preventable complications for patients, which may increase treatment costs.
Flagler’s AI project involved uploading clinical,
demographic, billing, and surgical information to the AyasdiAI platform, which then
employed machine learning to analyze the data and identify trends. Flagler’s
physicians are now provided with a fuller picture of their patients’ conditions,
which helps identify patients at highest risk, ensuring timely interventions that
produce positive outcomes and lower costs.
How Symphony AyasdiAI Works
The AyasdiAI application utilizes a category of mathematics called topological data analysis (TDA) to cluster similar patients together and locate parallels between those groups. “We then have the AI tools generate a carepath from this group, showing all events which should occur in the emergency department, at admission, and throughout the hospital stay,” Sanders told Healthcare IT News. “These events include all medications, diagnostic tests, vital signs, IVs, procedures and meals, and the ideal timing for the occurrence of each so as to replicate the results of this group.”
Caregivers then examine the data to determine the optimal
plan of care for each patient. Cost savings are figured into the overall
equation when choosing a treatment plan.
Flagler first used the AI program to examine trends among their pneumonia patients. They determined that nebulizer treatments should be started as soon as possible with pneumonia patients who also have chronic obstructive pulmonary disease (COPD).
“Once we have the data loaded, we use [an] unsupervised
learning AI algorithm to generate treatment groups,” Sanders told Healthcare
IT News. “In the case of our pneumonia patient data, Ayasdi produced nine
treatments groups. Each group was treated similarly, and statistics were given
to us to understand that group and how it differed from the other groups.”
Armed with this information, the hospital achieved an 80% greater physician adherence to order sets for pneumonia patients. This resulted in a savings of $1,350 per patient and reduced the readmission rates for pneumonia patients from 2.9% to 0.4%, reported Modern Healthcare.
The development of a machine-learning platform designed to
reduce variation in care (by helping physicians become more consistent at
following accepted clinical care guidelines) can be considered a warning shot
across the bow of the pathology profession.
This is a system that has the potential to become interposed
between the pathologist in the medical laboratory and the physicians who refer
specimens to the lab. Were that to happen, the deep experience and knowledge
that have long made pathologists the “doctor’s doctor” will be bypassed.
Physicians will stop making that first call to their pathologists, clinical
chemists, and laboratory scientists to discuss a patient’s condition and
consult on which test to order, how to interpret the results, and get guidance
on selecting therapies and monitoring the patient’s progress.
Instead, a “smart software solution” will be inserted into
the clinical workflow of physicians. This solution will automatically guide the
physician to follow the established care protocol. In turn, this will give the
medical laboratory the simple role of accepting a lab test order, performing
the analysis, and reporting the results.
If this were true, then it could be argued that a laboratory
test is a commodity and hospitals, physicians, and payers would argue that they
should buy these commodity lab tests at the cheapest price.
The new method employs a pH sensitive dye and AI algorithms to ‘distinguish between cells originating from normal and cancerous tissue, as well as among different types of cancer’ the researchers said
Might a pH-sensitive dye in tandem with an image analysis solution soon be used to identify cancerous cells within blood samples as well within tissue? Recent research indicates that could be a possibility. If further studies and clinical trials confirm this capability, then anatomic pathologists could gain another valuable tool to use in diagnosing cancers and other types of disease.
Currently, surgical pathologists use a variety of hematoxylin and eosin stains (H/E) to bring out useful features in cells and cell structures. So, staining tissue on glass slides is a common practice. Now, thanks to machine learning and artificial intelligence, anatomic pathologists may soon have a similar tool for spotting cancer cells within both tissue and blood samples.
Researchers at the National University of Singapore (NUS) have developed a method for identifying cancer that uses a pH sensitive dye called bromothymol blue. The dye reacts to various levels of acidity in cancer cells by turning colors. “The pH inside cancer cells tends to be higher than that of healthy cells. This phenomenon occurs at the very early phases of cancer development and becomes amplified as it progresses,” Labroots reported.
In “Machine Learning Based Approach to pH Imaging and Classification of Single Cancer Cells,” published in the journal APL Bioengineering, the NUS researchers wrote, “Here, we leverage a recently developed pH imaging modality and machine learning-based single-cell segmentation and classification to identify different cancer cell lines based on their characteristic intracellular pH. This simple method opens up the potential to perform rapid noninvasive identification of living cancer cells for early cancer diagnosis and further downstream analyses.”
According to an NUS news release, the bromothymol blue dye is “applied onto patients’ cells” being held ex vivo in cell culture dishes. The dye’s color changes depending on the acidity level of the cancer cells it encounters. Microscopic images of the now-visible cancers cells are taken, and a machine-learning algorithm analyzes the images before generating a report for the anatomic pathologist.
The NUS researchers claim the test can provide answers in about half an hour with 95% accuracy, Labroots reported.
AI Cell Analysis versus Laborious Medical Laboratory Steps
By developing an AI-driven method, Professor Lim and the NUS team sought to improve upon time-consuming techniques for identifying cells that traditionally involve using florescent probes, nanoparticles, and labeling steps, or for cells to be fixed or terminated.
“Unlike other cell analysis techniques, our approach uses simple, inexpensive equipment, and does not require lengthy preparation and sophisticated devices. Using AI, we are able to screen cells faster and accurately,” Professor Lim told Labroots. “Furthermore, we can monitor and analyze living cells without causing any toxicity to the cells or the need to kill them.”
The new technique may have implications for cancer detection in tumor tissue as well as in liquid biopsies.
“We are also exploring the possibility of performing the real-time analysis on circulating cancer cells suspended in blood,” Professor Lim said in the NUS news release. “One potential application for this would be in liquid biopsy where tumor cells that escaped from a primary tumor can be isolated in a minimally-invasive fashion from bodily fluids such as blood.”
Diagnosing Cancer in Real Time
The NUS’ method requires more research and clinical studies before it could become an actual tool for anatomic pathologists and other cancer diagnosticians. Additionally, the NUS researchers acknowledged that the focus on only four cell lines (normal cells, benign breast tumor cells, breast cancer cells, and pancreatic cancer cells) limited their study, as did lack of comparison with conventional florescent pH indicators.
Still, the NUS scientists are already planning more studies to advance their concept to different stages of cell malignancy. They envision a “real-time” version of the technique to enable recognition of cells and fast separation of those that need to be referred to clinical laboratories for molecular testing and/or genetic sequencing.
Medical laboratory leaders may want to follow the NUS study. An inexpensive AI-driven method that can accurately detect and classify cancer cells based on pH within the cells is provocative and may be eventually become integrated with other cancer diagnostics.