AI-Powered Diagnostic Tool Predicts Pancreatic Cancer Up to 3 Years in Advance

Given how difficult pancreatic cancer can be to diagnose, the tool could be life-saving.

AI-Powered Diagnostic Tool Predicts Pancreatic Cancer Up to 3 Years in Advance

By Adrianna Nine  in Extremetech

Pancreatic cancer might be relatively rare, but it’s one of the world’s leading causes of cancer-related death. Thanks to the organ’s placement deep within the abdomen, tumors on the pancreas can be difficult to detect, forcing doctors to rely on expensive and invasive imaging or blood tests instead. If these strategies are implemented too late, a patient’s odds of successfully completing treatment are slim.

A new tool powered by artificial intelligence might make early detection easier and more accessible. Researchers from Harvard Medical School and the University of Copenhagen partnered with the VA Boston Healthcare System and the Dana-Farber Cancer Institute to build a screening program that scans people’s medical records to determine their likelihood of developing pancreatic cancer. According to a paper published Tuesday in Nature Medicine, the tool can predict future pancreatic cancer diagnosis up to three years ahead with 88% accuracy.

The researchers trained an AI model on roughly 9 million patients’ health records obtained from the Danish National Patient Registry (DNPR) and the US Veterans Affairs database (US-VA). Both of these databases were used to train the model separately. Of the 6 million patients in the DNPR, 23,985 had been diagnosed with pancreatic cancer; of the 3 million US-VA patients, 3,900 had received the same diagnosis. Using this data, the AI model was trained to pick up on disease codes, comorbidities, and timelines that might indicate a future pancreatic cancer diagnosis—even if the disease codes themselves had nothing to do with the pancreas. ... ' 

Arterys: The future of precision medicine

Brought to my Attention , aimed at healthcare practices...

Arterys:  The future of precision medicine   that only human + AI can achieve. 

The Arterys platform extracts actionable insights from medical images to add clinical value, improve diagnostic decision making, efficiency and productivity.

Arterys is the medical imaging AI platform allowing you to weave leading AI clinical applications directly into your existing PACS or EHR driven workflow to make it a natural extension of what you already do.

We are making AI real by improving physician experience, accuracy of diagnosis and treatment, financial performance and outcomes that matter to patients and providers.

Accessible anywhere from any validated device via the cloud for faster performance, ease of deployment with no PHI exchange and completely secure.

See how Arterys is transforming healthcare through deep learning and AI

History:  It started at Stanford

Before we founded Arterys, we were graduate students, searching for a way to make medicine better. Faster. More precise.

We pushed the boundaries of our disciplines until they finally overlapped, intersecting cloud computing with cutting edge medical image acquisition.

Soon after, Arterys was founded on our shared beliefs:

Faster, smarter diagnosis where it counts the most

We started by improving diagnoses for newborns and kids with heart defects. At the time, pediatric cardiovascular disease was diagnosed with ultrasound, which only offers a partial view of the heart with no blood flow quantifications. Or an MRI captured over grueling hours on the scanning table.

Without accurate quantification, cardiac medicine was an educated guess with a high error rate.

The solution was obvious: 4D Flow technology to visualize and quantify blood flow in mere minutes.

But image archiving servers in hospitals couldn’t read 4D Flow’s big data files. So we applied a system that could: cloud computer processing.

With cloud computing, we could put those life-saving 4D Flow images into radiologists hands. In mere moments, physicians could diagnose and make accurate treatment decisions.

Anyone with access to a web browser could access Arterys to quantify regurgitant flow and determine if a child with heart defects needed surgery.

We were helping to save kids’ lives with better diagnoses.

But we weren’t satisfied.

Even though we put 4D Flow at radiologists’ fingertips, we still saw physicians manually drawing contours to quantify the size of cardiac ventricles.

Artificial Intelligence gets real

We decided to power Deep Learning AI with cloud computing GPUs to automatically quantify and segment ventricles as accurately as manual measurements by experienced physicians. In 2017, our technology received the first ever US FDA clearance for leveraging cloud computing and deep learning in a clinical setting.

But we were still not satisfied.

Physicians around the world approached us, asking if we could apply the same AI cloud computing approach to image processing and analysis to cancer patients. Then to liver patients. Then lung, breast, brain, and everything in between.

The word was getting out that radiologists could receive automatic accurate measurements through deep learning powered in the cloud.

All of your data, all in one place. Simple.

Despite our breakthroughs, existing radiology workflows put AI image acquisition and analysis out of reach. Working outside of a unified platform, physicians switched interfaces multiple times an hour to access different AI imaging tools. This stole vital time that could be spent with a patient. It was time to transcend silos with one united AI web platform.

Soon after, we received FDA clearance for our Arterys web-based AI imaging platform.

Despite these successes, we are still not satisfied.

We want to push medicine to make it even faster and more precise. We’ve just gotten started  .... ..'

Diagnosing Windows Internet Connections

I am sure you have all run into this. Just did and played with the router and wifi box for some time.  Then found it very helpful. But the internet diagnosis capability, which runs forever, does not execute these or suggest that you do?  Seems the 'skill' needs to improved.

Best Windows 10 commands to diagnose your Internet connection  By Lawrence Abrams in BleepingComputer

Diagnosing an Internet connection in Windows can be a frustrating experience. To help with this, we have outlined six Windows 10 commands that you can use to troubleshoot your Internet connection and help determine why you cannot reach a website.

All of the commands we list below require you to execute them via the Windows 10 command prompt.... " 

Models for Diagnosing Covid Diagnosis

Was wondering if validated ML models existed, had not heard of any.  What is missing?   The article suggests lack of data, can that be?    Yes, if the needed validated supporting metadata was not gathered systematically.   Passed this piece on to some people with clinical diagnosis experience. 

ML Models for Diagnosing Covid-19 Not Yet Suitable for Clinical Use

University of Cambridge (U.K.),  March 15, 2021

A review of studies containing descriptions of machine learning (ML) models for diagnosing Covid-19 by researchers at the U.K.'s University of Cambridge concluded that none are yet suitable for detecting or diagnosing the virus from standard medical imaging. The team ultimately reviewed 62 studies, and invalidated each model's suitability due to biases in study design, methodological flaws, lack of reproducibility, and publicly available "Frankenstein datasets." Many ML models were trained on sample datasets that were too small to be effective, failed to specify their data's origins, were trained and tested on the same data, or lacked involvement from radiologists and clinicians. Cambridge's James Rudd said, "These early studies show promise, but they suffer from a high prevalence of deficiencies in methodology and reporting, with none of the literature we reviewed reaching the threshold of robustness and reproducibility essential to support use in clinical practice."

Will AI Save Lives?

More on the healthcare uses pf AI methods for accurate diagnosis:

It's too soon to tell if DeepMind's medical AI will save any lives
Artificial intelligence trained on health records can now detect kidney injury up to two days before it occurs. The idea is that an advance warning could help doctors intervene earlier to prevent irreversible damage to the kidneys.

AIs are already touted as rivals to doctors when it comes to detecting medical conditions such as certain cancers and childhood illnesses. But few undergo rigorous clinical trials, so it’s still too early to know whether they are effective in practice.

Nenad Tomašev at DeepMind and colleagues trained an algorithm to predict the likelihood that a person who was admitted to hospital would go on to develop acute kidney injury (AKI).

They trained the AI using de-identified electronic health records from 703,782 US veterans aged between 18 and 90, who were admitted to hospital between October 2011 and September 2015.

Read more: AIs that diagnose diseases are starting to assist and replace doctors

AKI results in a dramatic drop in the rate at which the kidneys filter blood. This causes a decrease in urine production and a build-up of waste products in the blood, such as creatinine, a by-product of muscle breakdown. Both of these are used as measures for diagnosis.

Based on creatinine levels from a patient’s medical records, at a given time point the AI predicted whether a kidney injury would occur within the next 48 hours. Its accuracy was confirmed by comparing the prediction to whether the patient was later diagnosed.

The algorithm was fairly accurate at predicting the most severe forms of AKI. It correctly predicted 90 per cent of the cases in which the patient’s kidney function deteriorated so severely that they eventually required long-term dialysis.

It is difficult for doctors to anticipate kidney injury, so that level of accuracy is significant given the consequences of a severe injury, which include death or the need for a kidney transplant, says Eric Topol at Scripps Research in the US, who was not involved in the research.

However, the algorithm was far less accurate for all forms of AKI, correctly predicting only 55.8 of all episodes, with a ratio of two false alerts for one correct prediction. .... "

On the Trust of AI

Over time this will change, but mere statistical statements of performance are still hard to win with.  Augmented a doctor's reasoning capability is still a more powerful argument.   Note the argument of 'uniqueness' of their case mentioned here.   The doctor is still there to address unique cases.

AI Can Outperform Doctors. So Why Don’t Patients Trust It?   By Chiara Longoni,  Carey K. Morewedge in HBR

Our recent research indicates that patients are reluctant to use health care provided by medical artificial intelligence even when it outperforms human doctors. Why? Because patients believe that their medical needs are unique and cannot be adequately addressed by algorithms. To realize the many advantages and cost savings that medical AI promises, care providers must find ways to overcome these misgivings.

Medical artificial intelligence (AI) can perform with expert-level accuracy and deliver cost-effective care at scale. IBM’s Watson diagnoses heart disease better than cardiologists do. Chatbots dispense medical advice for the United Kingdom’s National Health Service in lieu of nurses. Smartphone apps now detect skin cancer with expert accuracy. Algorithms identify eye diseases just as well as specialized physicians. Some forecast that medical AI will pervade 90% of hospitals and replace as much as 80% of what doctors currently do. But for that to come about, the health care system will have to overcome patients’ distrust of AI. .... "

Wearable Sweat Sensors

Recall some research using sweat as a means of diagnosing.

Wearable sensors detect what's in your sweat
by Kara Manke, University of California - Berkeley

Needle pricks not your thing? A team of scientists at the University of California, Berkeley, is developing wearable skin sensors that can detect what's in your sweat.

They hope that one day, monitoring perspiration could bypass the need for more invasive procedures like blood draws, and provide real-time updates on health problems such as dehydration or fatigue.

In a paper appearing today in Science Advances, the team describes a new sensor design that can be rapidly manufactured using a "roll-to-roll" processing technique that essentially prints the sensors onto a sheet of plastic like words on a newspaper.

They used the sensors to monitor the sweat rate, and the electrolytes and metabolites in sweat, from volunteers who were exercising, and others who were experiencing chemically induced perspiration.

"The goal of the project is not just to make the sensors but start to do many subject studies and see what sweat tells us—I always say 'decoding' sweat composition," said Ali Javey, a professor of electrical engineering and computer science at UC Berkeley and senior author on the paper.

"For that we need sensors that are reliable, reproducible, and that we can fabricate to scale so that we can put multiple sensors in different spots of the body and put them on many subjects," said Javey, who also serves as a faculty scientist at Lawrence Berkeley National Laboratory. .....  "

Self Taught Computer Makes Diagnoses

We followed work ongoing at Stanford on pulmonary diagnosis.  These new methods and directions are interesting. Here diagnoses derived from X-Rays.

How Can Doctors Be Sure A Self-Taught Computer Is Making The Right Diagnosis?    By NPR 

Researchers at Stanford University have developed an algorithm that can use data from more than 100,000 chest X-rays, each tagged with a condition the person had been diagnosed with, to teach itself how to diagnose conditions linked to the images.

The algorithm had no guidance about what to look for; its only job was to teach itself by search for patterns, using deep learning.

Now the researchers are using a similar method to diagnose tuberculosis among HIV-positive patients in South Africa, hoping the program will help fill an urgent medical need.

Currently, the algorithm gets the diagnosis right 75% of the time, while human doctors are correct about 62% of the time.

Says Stanford researcher Matthew Lungren, "The ultimate thought from our group is that if we can combine the best of what humans offer in their diagnostic work and the best of what these models can offer, I think you're going to have a better level of health care for everybody."  .... ' 

Predictive Maintenance

Some of the earliest examples of useful AI were in the area of maintenance, and predictive diagnosis.  We examined a number. of engineering applications.  The market value of use is enormous.  From cars to jet engines.   Natural to look at the latest machine learning techniques here.  MIT's progress in the area.   Detailed papers are pointed to.

Let your car tell you what it needs
MIT team develops software that can tell if tires need air, spark plugs are bad, or air filter needs replacing.

David L. Chandler | MIT News Office 

Imagine hopping into a ride-share car, glancing at your smartphone, and telling the driver that the car’s left front tire needs air, its air filter should be replaced next week, and its engine needs two new spark plugs.

Within the next year or two, people may be able to get that kind of diagnostic information in just a few minutes, in their own cars or any car they happen to be in. They wouldn’t need to know anything about the car’s history or to connect to it in any way; the information would be derived from analyzing the car’s sounds and vibrations, as measured by the phone’s microphone and accelerometers.

The MIT research behind this idea has been reported in a series of papers, most recently in the November issue of the journal Engineering Applications of Artificial Intelligence. The new paper’s co-authors include research scientist Joshua Siegel PhD ’16; Sanjay Sarma, the Fred Fort Flowers and Daniel Fort Flowers Professor of Mechanical Engineering and vice president of open learning at MIT; and two others.  .....  " 

Disease Diagnosis, Data Analysis

Instructive, technical piece on disease diagnosis using Bayesian analysis.  Includes R code for implementation.  Cautionary analysis of the results.  Particular attention paid to false alarm rate. Useful for analyzing many kinds of data.

Health Scanning with Sweat Sensors

From our local University of Cincinnati, whose board I was on for years:

" ... A University of Cincinnati partnership is reporting a significant leap forward in health diagnostics that are more accurate than what’s available today, yet considerably less invasive than something like a blood screening. It’s a lightweight, wearable device that analyzes sweat by using a smartphone.  UC researcher Jason Heikenfeld is holding next generation of sweat sensor pad and flexible bluetooth circuit. .... 

In an article published in the latest issue of IEEE Spectrum Magazine, Jason Heikenfeld, a UC professor of electrical engineering and computing systems, highlights new developments regarding the gadget that can gather vital medical information in almost real time.  ... " .    Bravo. Look forward to hearing more.

Smart Digital Diapers go Mobile

In the NYT: A Digital Diaper that tracks infants health.   Reminds me of the 'smart toilet' idea pioneered in Japan.  Had not heard this specifically talked about in R&D in the diaper space.  The connection to a smartphone App makes it an example of a remote heath monitor.    " .... A New York start-up called Pixie Scientific has developed a diaper that the company says can detect possible urinary tract infections, kidney dysfunctions, and dehydration, accompanied by a smartphone app that can transmit the information to a physician. ... "  Also in Venturebeat.

IBM's Watson Does Diagnoses

Another piece about uses for the IBM Watson Computer.  Once again, patient diagnosis was an early application for artificial intelligence.  In particular pulmonary diagnosis was used as an early example.  Such systems can be seen as advisers to physician.   It makes sense, there is gathered personal data and a large library of patterns that resulted in specific diagnoses.  It is straightforward to find specific matches and attribute to each a level of credibility.   Could work where there are large quantities of complex data to be matched.