In an initial study, Canary analyzed speech recordings with AI and identified early stage Alzheimer’s with 96 percent accuracy.
Dave Arnold retired in his 60s and began spending time volunteering in local schools. But then he started misplacing items, forgetting appointments and losing his sense of direction. Eventually he was diagnosed with early stage Alzheimer’s.
“Hearing the diagnosis made me very emotional and tearful,” he said. “I immediately thought of all my mom had experienced.” His mother suffered with the condition for years before passing away. Over the last year, Arnold has worked for the Alzheimer’s Association as one of its early stage advisors, sharing his insights to help others in the initial stages of the disease.
Arnold was diagnosed sooner than many others. It's important to find out early, when interventions can make the most difference. One promising avenue is looking at how people talk. Research has shown that Alzheimer’s affects a part of the brain that controls speech, resulting in small changes before people show other signs of the disease.
Now, Canary Speech, a company based in Utah, is using AI to examine elements like the pitch of a person’s voice and their pauses. In an initial study, Canary analyzed speech recordings with AI and identified early stage Alzheimer’s with 96 percent accuracy.
Developing the AI model
Canary Speech’s CEO, Henry O’Connell, met cofounder Jeff Adams about 40 years before they started the company. Back when they first crossed paths, they were both living in Bethesda, Maryland; O’Connell was a research fellow at the National Institutes of Health studying rare neurological diseases, while Adams was working to decode spy messages. Later on, Adams would specialize in building mathematical models to analyze speech and sound as a team leader in developing Amazon's Alexa.
It wasn't until 2015 that they decided to make use of the fit between their backgrounds. ““We established Canary Speech in 2017 to build a product that could be used in multiple languages in clinical environments,” O'Connell says.
The need is growing. About 55 million people worldwide currently live with Alzheimer’s, a number that is expected to double by 2050. Some scientists think the disease results from a buildup of plaque in the brain. It causes mild memory loss at first and, over time, this issue get worse while other symptoms, such as disorientation and hallucinations, can develop. Treatment to manage the disease is more effective in the earlier stages, but detection is difficult since mild symptoms are often attributed to the normal aging process.
O’Connell and Adams specialize in the complex ways that Alzheimer’s effects how people speak. Using AI, their mathematical model analyzes 15 million data points every minute, focusing on certain features of speech such as pitch, pauses and elongation of words. It also pays attention to how the vibrations of vocal cords change in different stages of the disease.
To create their model, the team used a type of machine learning called deep neural nets, which looks at multiple layers of data - in this case, the multiple features of a person’s speech patterns.
“Deep neural nets allow us to look at much, much larger data sets built out of millions of elements,” O’Connell explained. “Through machine learning and AI, we’ve identified features that are very sensitive to an Alzheimer’s patient versus [people without the disease] and also very sensitive to mild cognitive impairment, early stage and moderate Alzheimer's.” Based on their learnings, Canary is able to classify the disease stage very quickly, O’Connell said.
“When we’re listening to sublanguage elements, we’re really analyzing the direct result of changes in the brain in the physical body,” O’Connell said. “The brain controls your vocal cords: how fast they vibrate, the expansion of them, the contraction.” These factors, along with where people put their tongues when talking, function subconsciously and result in subtle changes in the sounds of speech.
Further testing is needed
In an initial trial, Canary analyzed speech recordings from phone calls to a large U.S. health insurer. They looked at the audio recordings of 651 policyholders who had early stage Alzheimer’s and 1018 who did not have the condition, aiming for a representative sample of age, gender and race. They used this data to create their first diagnostic model and found that it was 96 percent accurate in identifying Alzheimer’s.
Christian Herff, an assistant professor of neuroscience at Maastricht University in the Netherlands, praised this approach while adding that further testing is needed to assess its effectiveness.
“I think the general idea of identifying increased risk for cognitive impairment based on speech characteristics is very feasible, particularly when change in a user’s voice is monitored, for example, by recording speech every year,” Herff said. He noted that this can only be a first indication, not a full diagnosis. The accuracy still needs to be validated in studies that follows individuals over a period of time, he said.
Toby Walsh, a professor of artificial intelligence at the University of New South Wales, also thinks Canary’s tool has potential but highlights that Canary could diagnose some people who don’t really have the disease. “This is an interesting and promising application of AI,” he said, “but these tools need to be used carefully. Imagine the anxiety of being misdiagnosed with Alzheimer’s.”
As with many other AI tools, privacy and bias are additional issues to monitor closely, Walsh said.
Other languages
A related issue is that not everyone is fluent in English. Mahnaz Arvaneh, a senior lecturer in automatic control and systems engineering at the University of Sheffield, said this could be a blind spot.
“The system may not be very accurate for those who have English as their second language as their speaking patterns would be different, and any issue might be because of language deficiency rather than cognitive issues,” Arvaneh said.
The team is expanding to multiple languages starting with Japanese and Spanish. The elements of the model that make up the algorithm are very similar, but they need to be validated and retrained in a different language, which will require access to more data.
Recently, Canary analyzed the phone calls of 233 Japanese patients who had mild cognitive impairment and 704 healthy people. Using an English model they were able to identify the Japanese patients who had mild cognitive impairment with 78 percent accuracy. They also developed a model in Japanese that was 45 percent accurate, and they’re continuing to train it with more data.
The future
Canary is using their model to look at other diseases like Huntington’s and Parkinson’s. They’re also collaborating with pharmaceuticals to validate potential therapies for Alzheimer’s. By looking at speech patterns over time, Canary can get an indication of how well these drugs are working.
Dave Arnold and his wife dance at his nephew’s wedding in Rochester, New York, ten years ago, before his Alzheimer's diagnosis.
Dave Arnold
Ultimately, they want to integrate their tool into everyday life. “We want it to be used in a smartphone, or a teleconference call so that individuals could be examined in their home,” O’Connell said. “We could follow them over time and work with clinical teams and hospitals to improve the evaluation of patients and contribute towards an accurate diagnosis.”
Arnold, the patient with early stage Alzheimer’s, sees great promise. “The process of getting a diagnosis is already filled with so much anxiety,” he said. “Anything that can be done to make it easier and less stressful would be a good thing, as long as it’s proven accurate.”
Marielle Gross, a professor at the University of Pittsburgh, shows patients a new app that tracks how their samples are used during biomedical research.
Tissue donors can track what scientists do with their samples while safeguarding privacy, through a pilot program initiated in October by researchers at the Johns Hopkins Berman Institute of Bioethics and the University of Pittsburgh’s Institute for Precision Medicine. The program uses blockchain technology to offer patients this opportunity through the University of Pittsburgh's Breast Disease Research Repository, while assuring that their identities remain anonymous to investigators.
A blockchain is a digital, tamper-proof ledger of transactions duplicated and distributed across a computer system network. Whenever a transaction occurs with a patient’s sample, multiple stakeholders can track it while the owner’s identity remains encrypted. Special certificates called “nonfungible tokens,” or NFTs, represent patients’ unique samples on a trusted and widely used blockchain that reinforces transparency.
Blockchain could be used to notify people if cancer researchers discover that they have certain risk factors.
“Healthcare is very data rich, but control of that data often does not lie with the patient,” said Julius Bogdan, vice president of analytics for North America at the Healthcare Information and Management Systems Society (HIMSS), a Chicago-based global technology nonprofit. “NFTs allow for the encapsulation of a patient’s data in a digital asset controlled by the patient.” He added that this technology enables a more secure and informed method of participating in clinical and research trials.
Without this technology, de-identification of patients’ samples during biomedical research had the unintended consequence of preventing them from discovering what researchers find -- even if that data could benefit their health. A solution was urgently needed, said Marielle Gross, assistant professor of obstetrics, gynecology and reproductive science and bioethics at the University of Pittsburgh School of Medicine.
“A researcher can learn something from your bio samples or medical records that could be life-saving information for you, and they have no way to let you or your doctor know,” said Gross, who is also an affiliate assistant professor at the Berman Institute. “There’s no good reason for that to stay the way that it is.”
For instance, blockchain could be used to notify people if cancer researchers discover that they have certain risk factors. Gross estimated that less than half of breast cancer patients are tested for mutations in BRCA1 and BRCA2 — tumor suppressor genes that are important in combating cancer. With normal function, these genes help prevent breast, ovarian and other cells from proliferating in an uncontrolled manner. If researchers find mutations, it’s relevant for a patient’s and family’s follow-up care — and that’s a prime example of how this newly designed app could play a life-saving role, she said.
Liz Burton was one of the first patients at the University of Pittsburgh to opt for the app -- called de-bi, which is short for decentralized biobank -- before undergoing a mastectomy for early-stage breast cancer in November, after it was diagnosed on a routine mammogram. She often takes part in medical research and looks forward to tracking her tissues.
“Anytime there’s a scientific experiment or study, I’m quick to participate -- to advance my own wellness as well as knowledge in general,” said Burton, 49, a life insurance service representative who lives in Carnegie, Pa. “It’s my way of contributing.”
Liz Burton was one of the first patients at the University of Pittsburgh to opt for the app before undergoing a mastectomy for early-stage breast cancer.
Liz Burton
The pilot program raises the issue of what investigators may owe study participants, especially since certain populations, such as Black and indigenous peoples, historically were not treated in an ethical manner for scientific purposes. “It’s a truly laudable effort,” Tamar Schiff, a postdoctoral fellow in medical ethics at New York University’s Grossman School of Medicine, said of the endeavor. “Research participants are beautifully altruistic.”
Lauren Sankary, a bioethicist and associate director of the neuroethics program at Cleveland Clinic, agrees that the pilot program provides increased transparency for study participants regarding how scientists use their tissues while acknowledging individuals’ contributions to research.
However, she added, “it may require researchers to develop a process for ongoing communication to be responsive to additional input from research participants.”
Peter H. Schwartz, professor of medicine and director of Indiana University’s Center for Bioethics in Indianapolis, said the program is promising, but he wonders what will happen if a patient has concerns about a particular research project involving their tissues.
“I can imagine a situation where a patient objects to their sample being used for some disease they’ve never heard about, or which carries some kind of stigma like a mental illness,” Schwartz said, noting that researchers would have to evaluate how to react. “There’s no simple answer to those questions, but the technology has to be assessed with an eye to the problems it could raise.”
To truly make a difference, blockchain must enable broad consent from patients, not just de-identification.
As a result, researchers may need to factor in how much information to share with patients and how to explain it, Schiff said. There are also concerns that in tracking their samples, patients could tell others what they learned before researchers are ready to publicly release this information. However, Bogdan, the vice president of the HIMSS nonprofit, believes only a minimal study identifier would be stored in an NFT, not patient data, research results or any type of proprietary trial information.
Some patients may be confused by blockchain and reluctant to embrace it. “The complexity of NFTs may prevent the average citizen from capitalizing on their potential or vendors willing to participate in the blockchain network,” Bogdan said. “Blockchain technology is also quite costly in terms of computational power and energy consumption, contributing to greenhouse gas emissions and climate change.”
In addition, this nascent, groundbreaking technology is immature and vulnerable to data security flaws, disputes over intellectual property rights and privacy issues, though it does offer baseline protections to maintain confidentiality. To truly make a difference, blockchain must enable broad consent from patients, not just de-identification, said Robyn Shapiro, a bioethicist and founding attorney at Health Sciences Law Group near Milwaukee.
The Henrietta Lacks story is a prime example, Shapiro noted. During her treatment for cervical cancer at Johns Hopkins, Lacks’s tissue was de-identified (albeit not entirely, because her cell line, HeLa, bore her initials). After her death, those cells were replicated and distributed for important and lucrative research and product development purposes without her knowledge or consent.
Nonetheless, Shapiro thinks that the initiative by the University of Pittsburgh and Johns Hopkins has potential to solve some ethical challenges involved in research use of biospecimens. “Compared to the system that allowed Lacks’s cells to be used without her permission, Shapiro said, “blockchain technology using nonfungible tokens that allow patients to follow their samples may enhance transparency, accountability and respect for persons who contribute their tissue and clinical data for research.”
Read more about laws that have prevented people from the rights to their own cells.
The U.S. has the highest incarceration rate in the world, costing $182 billion per year, partly because its antiquated data systems often fail to identify people who should be released. A tech nonprofit is trying to change that.
Recidiviz has built a platform that connects the different databases, with the goal of identifying people who are already qualified for release but remain behind bars or on supervision. “Think of Recidiviz like Google Maps,” says Jacoby, who worked on Maps when she was at the tech giant. Google Maps takes in data from different sources – satellite images, street maps, local business data — and organizes it into one easy view. “Recidiviz does something similar with criminal justice data,” Jacoby explains, “making it easy to identify people eligible to come home or to move to less intensive levels of supervision.”
People like Jacoby’s uncle. His experience with incarceration is what inspired her passion for criminal justice reform in the first place.
The problems are vast
The U.S. has the highest incarceration rate in the world — 2 million people according to the watchdog group, Prison Policy Initiative — at a cost of $182 billion a year. The numbers could be a lot lower if not for an array of problems including inaccurate sentencing calculations, flawed algorithms and parole violations laws.
Sentencing miscalculations
To determine eligibility for release, the current system requires corrections officers to check 21 different requirements spread across five different databases for each of the 90 to 100 people under their supervision. These manual calculations are time prohibitive, says Jacoby, and fall victim to human error.
In addition, Recidiviz found that policies aimed at helping to reduce the prison population don’t always work correctly. A key example is time off for good behavior laws that allow inmates to earn one day off for every 30 days of good behavior. Some states' data systems are built to calculate time off as one day per month of good behavior, rather than per day. Over the course of a decade-long sentence, Jacoby says these miscalculations can lead to a huge discrepancy in the calculated release data and the actual release date.
Algorithms
Commercial algorithm-based software systems for risk assessment continue to be widely used in the criminal justice system, even though a 2018 study published in Science Advances exposed their limitations. After the study went viral, it took three years for the Justice Department to issue a report on their own flawed algorithms used to reduce the federal prison population as part of the 2018 First Step Act. The program, it was determined, overestimated the risk of putting inmates of color into early-release programs.
Despite its name, Recidiviz does not build these types of algorithms for predicting recidivism, or whether someone will commit another crime after being released from prison. Rather, Jacoby says the company’s "descriptive analytics” approach is specifically intended to weed out incarceration inequalities and avoid algorithmic pitfalls.
Parole violation laws
Research shows that 350,000 people a year — about a quarter of the total prison population — are sent back not because they’ve committed another crime, but because they’ve broken a specific rule of their probation. “Things that wouldn't send you or I to prison, but would send someone on parole,” such as crossing county lines or being in the presence of alcohol when they shouldn’t be, are inflating the prison population, says Jacoby.
It’s personal for the co-founder and CEO
“I grew up with an uncle who went into the prison system,” Jacoby says. At 19, he was sentenced to ten years in prison for a non-violent crime. A few months after being released from jail, he was sent back for a non-violent parole violation.
“For my family, the fact that one in four prison admissions are driven not by a crime but by someone who's broken a rule on probation and parole was really profound because that happened to my uncle,” Jacoby says. The experience led her to begin studying criminal justice in high school, then college. She continued her dive into how the criminal justice system works as part of her Passion Project while at Google, a program that allows employees to spend 20 percent of their time on pro-bono work. Two colleagues whose family members had also been stuck in the system joined her.
As part of the project, Jacoby interviewed hundreds of people involved in the criminal justice system. “Those on the right, those on the left, agreed that bad data was slowing down reform,” she says. Their research brought them to North Dakota where they began to understand the root of the problem. The corrections department is making “huge, consequential decisions every day [without] … the data,” Jacoby says. In a new video by Recidiviz not yet released, Jacoby recounts her exchange with the state’s director of corrections who told her, “‘It’s not that we have the data and we just don’t know how to make it public; we don’t have the information you think we have.'"
A mock-up (with fake data) of the types of dashboards and insights that Recidiviz provides to state governments.
Recidiviz
As a software engineer, Jacoby says the comment made no sense to her — until she witnessed it first-hand. “We spent a lot of time driving around in cars with corrections directors and parole officers watching them use these incredibly taxing, frankly terrible, old data systems,” Jacoby says.
As they weeded through thousands of files — some computerized, some on paper — they unearthed the consequences of bad data: Hundreds of people in prison well past their release date and thousands more whose release from parole was delayed because of minor paperwork issues. They found individuals stuck in parole because they hadn’t checked one last item off their eligibility list — like simply failing to provide their parole officer with a paystub. And, even when parolees advocated for themselves, the archaic system made it difficult for their parole officers to confirm their eligibility, so they remained in the system. Jacoby and her team also unpacked specific policies that drive racial disparities — such as fines and fees.
The Solution
It’s more than a trivial technical challenge to bring the incomplete, fragmented data onto a 21st century data platform. It takes months for Recidiviz to sift through a state’s information systems to connect databases “with the goal of tracking a person all the way through their journey and find out what’s working for 18- to 25-year-old men, what’s working for new mothers,” explains Jacoby in the video.
TED Talk: How bad data traps people in the U.S. justice system
TED Fellow Clementine Jacoby's TED Talk went live on Jan. 13. It describes how we can fix bad data in the criminal justice system, "bringing thousands of people home, reducing costs and improving public safety along the way."
Clementine Jacoby • TED2022
Ojmarrh Mitchell, an associate professor in the School of Criminology and Criminal Justice at Arizona State University, who is not involved with the company, says what Recidiviz is doing is “remarkable.” His perspective goes beyond academic analysis. In his pre-academic years, Mitchell was a probation officer, working within the framework of the “well known, but invisible” information sharing issues that plague criminal justice departments. The flexibility of Recidiviz’s approach is what makes it especially innovative, he says. “They identify the specific gaps in each jurisdiction and tailor a solution for that jurisdiction.”
On the downside, the process used by Recidiviz is “a bit opaque,” Mitchell says, with few details available on how Recidiviz designs its tools and tracks outcomes. By sharing more information about how its actions lead to progress in a given jurisdiction, Recidiviz could help reformers in other places figure out which programs have the best potential to work well.
The eleven states in which Recidiviz is working include California, Colorado, Maine, Michigan, Missouri, Pennsylvania and Tennessee. And a pilot program launched last year in Idaho, if scaled nationally, with could reduce the number of people in the criminal justice system by a quarter of a million people, Jacoby says. As part of the pilot, rather than relying on manual calculations, Recidiviz is equipping leaders and the probation officers with actionable information with a few clicks of an app that Recidiviz built.
Mitchell is disappointed that there’s even the need for Recidiviz. “This is a problem that government agencies have a responsibility to address,” he says. “But they haven’t.” For one company to come along and fill such a large gap is “remarkable.”