Article Type
Changed
Thu, 09/26/2024 - 11:35

Digital pathology assisted by artificial intelligence (AI) has the potential to transform the diagnosis and treatment of fibrotic liver disease in the next few years and to reshape clinical trials, clearing the way for new therapies.

Although the technology is not yet approved for routine clinical use, it’s constantly improving and aims to address the limitations inherent in today’s pathology processes.

“You do a biopsy, but instead of having a pathologist read it with their very rough scores of stage 1, 2, or 3, you read it by an AI-driven machine that can quantify it with a score of 1.5 or 1.75 instead of 1 or 2,” Vlad Ratziu, MD, PhD, professor of hepatology at the Sorbonne Université and Hôpital Pitié-Salpêtrière Medical School in Paris, France, and coeditor of The Journal of Hepatology, said in an interview.

“The technology is automated, more sensitive to change, and more highly quantitative. It has implications for liver disease diagnoses, clinical trials, and treatments,” added Dr. Ratziu, who has written about the promise and challenges inherent in developing treatments for metabolic dysfunction–associated steatotic liver disease (MASLD).

To explore the potential impact of AI-powered technologies for the clinic, this news organization spoke with representatives from three companies identified by Dr. Ratziu as leaders in the field: HistoIndexPathAI, and PharmaNest. Each company uses proprietary technology augmented by AI, and their tools have been used in published trials.
 

Moving Toward Better Diagnoses and Disease Management

The traditional approach for staging liver fibrosis relies on trained pathologists manually evaluating stained tissue samples obtained from biopsies of the liver.

But this method, though still considered the gold standard, doesn’t always provide the granularity needed for an accurate diagnosis or a reliable assessment in clinical trials, said Dean Tai, PhD, HistoIndex’s cofounder and chief scientific officer.

Although noninvasive tests (NITs), alone or with traditional histologic examination, are increasingly used during clinical management because they are less invasive and more repeatable for disease monitoring, they are limited in their precision and ability to provide comprehensive information, Dr. Tai said. That’s because “no single NIT or single-dimensional measurement of a biomarker offers a full assessment of disease activity, fibrogenic drive, and fibrosis load.”

In contrast, AI provides “a highly reproducible and objective assessment of liver fibrosis severity,” he said. “It eliminates the variability associated with staining methods, while revealing changes in the nano-architecture and morphology of collagen fibers not discernible by the human eye or current NITs, especially in the early stages of fibrosis or in cases of simultaneous progression and regression.”

Mathieu Petitjean, PhD, founder and CEO of PharmaNest, has a similar view. 

Although degree of liver fibrosis is associated with long-term outcomes of patients with MASLD, “poor detection thresholds due to their categorical nature mean that small and relevant changes are not reflected by changes in staging,” he said. “The reliable detection [with AI] of subtle changes in the phenotypes of fibrosis will significantly enrich the understanding of progression and regression of fibrosis severity.”

The ability of AI-based tools to see patterns the human eye cannot also means they could “help in predicting which patient may respond to a drug, in order to get the right treatments to the right patients as soon as possible,” said Katy Wack, PhD, vice president of clinical development at PathAI.

“Additionally, AI-based algorithms have been developed to provide more quantitative continuous scores to better capture change and discover new tissue-based biomarkers, which may be prognostic or predictive of clinical benefit,” she said. 

Such tools are currently undergoing testing and validation for use in trials and diagnostically.

The standardization and reproducibility offered by AI-driven technology could facilitate more consistent diagnoses across different healthcare settings, Dr. Tai suggested. “As the integration of the technology with other blood-, imaging-, and omics-based techniques evolves, it may enable earlier detection of liver diseases, more accurate monitoring of disease progression, and better evaluation of treatment responses, ultimately improving patient care and outcomes.”
 

 

 

More Effective Clinical Trials

The limitations of conventional pathology may be responsible, at least in part, for the repeated failure of novel compounds to move from phase 2 to phase 3 clinical trials, and from clinical trials to approval, the sources agreed.

“In clinical trials, patients are subject to enrollment criteria using liver biopsies, which are scored with a composite scoring system involving four different histologic components to grade and stage the disease,” Dr. Wack noted. 

However, there is wide variability between pathologists on biopsy scoring, and an individual pathologist presented with the same sample may give it a different score after some time has passed, she said.

That means “we are using a nonstandardized and inconsistent scoring system to determine whether a patient can be enrolled or not into a trial,” Dr. Wack said. 

The change in the composite score over a follow-up period, usually 1-2 years, determines whether a patient has responded to the candidate drug and, ultimately, whether that drug could be considered for approval, Dr. Wack said.

Because scores at the baseline and follow-up timepoints are not precise and inconsistent across pathologist readers, and even the same reader over time, there are often many “false-positive” and “false-negative” responses that can result in potential therapeutics either failing or succeeding in clinical trials, she said.

To address this variability in biopsy scoring as it relates to clinical trials, regulatory bodies have recommended a consensus approach, in which multiple pathologists read the same biopsy independently and a median score is used, or pathologists convene to come to an agreement, Dr. Wack said. 

“This is a very costly and burdensome approach and is still subject to interconsensus panel variation,” she said.

The introduction of digital pathology using validated digital viewers, where pathologists can view a glass slide digitally and pan and zoom over the image as they can with a microscope, means that many pathologists can read the same slide in parallel, she explained.

“If they need to discuss, they can do so efficiently over a phone call, each using their own computer screen and shared annotation tools to facilitate their discussion.”

Although this consensus approach can improve consistency, it still leads to variability in scoring across different groups of pathologists, Dr. Wack said.

This is where AI-assisted pathology comes into play.

“With this approach, a pathologist still views the image digitally, but an algorithm has predicted and highlighted key features and recommended quantitative scores,” she said.

This approach has been shown to increase precision for pathologists, thereby increasing reproducibility and standardizing scoring across timepoints and clinical trials.
 

What’s Ahead

These AI tools could address pathology’s lack of scalability, the result of a limited number of trained pathologists capable of doing liver biopsy assessments, Dr. Tai said. 

“Digital pathology workflows enable the transformation of conventional histologic glass slides into large digital images using scanners, allowing significant productivity gains in terms of workflow and collaboration,” he said.

Although AI-assisted pathology tools are still being validated, their promise for improving diagnoses and uncovering new treatments is clear, the interviewees agreed.

Extending its use to stage fibrosis in other liver diseases, such as primary biliary cholangitis, primary sclerosing cholangitis, and alcoholic liver disease, is also in progress on an experimental basis but will take time to validate.

“The landscape will evolve quickly in the coming 3-4 years,” Dr. Petitjean predicted. “To start, their intended use will likely be limited to a decision-support tool to enhance the performance of pathologists and perhaps stratify or triage cases sent for routine vs expert review.”

Dr. Petitjean even suggested that the increasing role of NITs and the amount of data being generated prospectively and retrospectively around liver biomarkers could mean that liver biopsies might not be needed one day.

A version of this article appeared on Medscape.com.

Publications
Topics
Sections

Digital pathology assisted by artificial intelligence (AI) has the potential to transform the diagnosis and treatment of fibrotic liver disease in the next few years and to reshape clinical trials, clearing the way for new therapies.

Although the technology is not yet approved for routine clinical use, it’s constantly improving and aims to address the limitations inherent in today’s pathology processes.

“You do a biopsy, but instead of having a pathologist read it with their very rough scores of stage 1, 2, or 3, you read it by an AI-driven machine that can quantify it with a score of 1.5 or 1.75 instead of 1 or 2,” Vlad Ratziu, MD, PhD, professor of hepatology at the Sorbonne Université and Hôpital Pitié-Salpêtrière Medical School in Paris, France, and coeditor of The Journal of Hepatology, said in an interview.

“The technology is automated, more sensitive to change, and more highly quantitative. It has implications for liver disease diagnoses, clinical trials, and treatments,” added Dr. Ratziu, who has written about the promise and challenges inherent in developing treatments for metabolic dysfunction–associated steatotic liver disease (MASLD).

To explore the potential impact of AI-powered technologies for the clinic, this news organization spoke with representatives from three companies identified by Dr. Ratziu as leaders in the field: HistoIndexPathAI, and PharmaNest. Each company uses proprietary technology augmented by AI, and their tools have been used in published trials.
 

Moving Toward Better Diagnoses and Disease Management

The traditional approach for staging liver fibrosis relies on trained pathologists manually evaluating stained tissue samples obtained from biopsies of the liver.

But this method, though still considered the gold standard, doesn’t always provide the granularity needed for an accurate diagnosis or a reliable assessment in clinical trials, said Dean Tai, PhD, HistoIndex’s cofounder and chief scientific officer.

Although noninvasive tests (NITs), alone or with traditional histologic examination, are increasingly used during clinical management because they are less invasive and more repeatable for disease monitoring, they are limited in their precision and ability to provide comprehensive information, Dr. Tai said. That’s because “no single NIT or single-dimensional measurement of a biomarker offers a full assessment of disease activity, fibrogenic drive, and fibrosis load.”

In contrast, AI provides “a highly reproducible and objective assessment of liver fibrosis severity,” he said. “It eliminates the variability associated with staining methods, while revealing changes in the nano-architecture and morphology of collagen fibers not discernible by the human eye or current NITs, especially in the early stages of fibrosis or in cases of simultaneous progression and regression.”

Mathieu Petitjean, PhD, founder and CEO of PharmaNest, has a similar view. 

Although degree of liver fibrosis is associated with long-term outcomes of patients with MASLD, “poor detection thresholds due to their categorical nature mean that small and relevant changes are not reflected by changes in staging,” he said. “The reliable detection [with AI] of subtle changes in the phenotypes of fibrosis will significantly enrich the understanding of progression and regression of fibrosis severity.”

The ability of AI-based tools to see patterns the human eye cannot also means they could “help in predicting which patient may respond to a drug, in order to get the right treatments to the right patients as soon as possible,” said Katy Wack, PhD, vice president of clinical development at PathAI.

“Additionally, AI-based algorithms have been developed to provide more quantitative continuous scores to better capture change and discover new tissue-based biomarkers, which may be prognostic or predictive of clinical benefit,” she said. 

Such tools are currently undergoing testing and validation for use in trials and diagnostically.

The standardization and reproducibility offered by AI-driven technology could facilitate more consistent diagnoses across different healthcare settings, Dr. Tai suggested. “As the integration of the technology with other blood-, imaging-, and omics-based techniques evolves, it may enable earlier detection of liver diseases, more accurate monitoring of disease progression, and better evaluation of treatment responses, ultimately improving patient care and outcomes.”
 

 

 

More Effective Clinical Trials

The limitations of conventional pathology may be responsible, at least in part, for the repeated failure of novel compounds to move from phase 2 to phase 3 clinical trials, and from clinical trials to approval, the sources agreed.

“In clinical trials, patients are subject to enrollment criteria using liver biopsies, which are scored with a composite scoring system involving four different histologic components to grade and stage the disease,” Dr. Wack noted. 

However, there is wide variability between pathologists on biopsy scoring, and an individual pathologist presented with the same sample may give it a different score after some time has passed, she said.

That means “we are using a nonstandardized and inconsistent scoring system to determine whether a patient can be enrolled or not into a trial,” Dr. Wack said. 

The change in the composite score over a follow-up period, usually 1-2 years, determines whether a patient has responded to the candidate drug and, ultimately, whether that drug could be considered for approval, Dr. Wack said.

Because scores at the baseline and follow-up timepoints are not precise and inconsistent across pathologist readers, and even the same reader over time, there are often many “false-positive” and “false-negative” responses that can result in potential therapeutics either failing or succeeding in clinical trials, she said.

To address this variability in biopsy scoring as it relates to clinical trials, regulatory bodies have recommended a consensus approach, in which multiple pathologists read the same biopsy independently and a median score is used, or pathologists convene to come to an agreement, Dr. Wack said. 

“This is a very costly and burdensome approach and is still subject to interconsensus panel variation,” she said.

The introduction of digital pathology using validated digital viewers, where pathologists can view a glass slide digitally and pan and zoom over the image as they can with a microscope, means that many pathologists can read the same slide in parallel, she explained.

“If they need to discuss, they can do so efficiently over a phone call, each using their own computer screen and shared annotation tools to facilitate their discussion.”

Although this consensus approach can improve consistency, it still leads to variability in scoring across different groups of pathologists, Dr. Wack said.

This is where AI-assisted pathology comes into play.

“With this approach, a pathologist still views the image digitally, but an algorithm has predicted and highlighted key features and recommended quantitative scores,” she said.

This approach has been shown to increase precision for pathologists, thereby increasing reproducibility and standardizing scoring across timepoints and clinical trials.
 

What’s Ahead

These AI tools could address pathology’s lack of scalability, the result of a limited number of trained pathologists capable of doing liver biopsy assessments, Dr. Tai said. 

“Digital pathology workflows enable the transformation of conventional histologic glass slides into large digital images using scanners, allowing significant productivity gains in terms of workflow and collaboration,” he said.

Although AI-assisted pathology tools are still being validated, their promise for improving diagnoses and uncovering new treatments is clear, the interviewees agreed.

Extending its use to stage fibrosis in other liver diseases, such as primary biliary cholangitis, primary sclerosing cholangitis, and alcoholic liver disease, is also in progress on an experimental basis but will take time to validate.

“The landscape will evolve quickly in the coming 3-4 years,” Dr. Petitjean predicted. “To start, their intended use will likely be limited to a decision-support tool to enhance the performance of pathologists and perhaps stratify or triage cases sent for routine vs expert review.”

Dr. Petitjean even suggested that the increasing role of NITs and the amount of data being generated prospectively and retrospectively around liver biomarkers could mean that liver biopsies might not be needed one day.

A version of this article appeared on Medscape.com.

Digital pathology assisted by artificial intelligence (AI) has the potential to transform the diagnosis and treatment of fibrotic liver disease in the next few years and to reshape clinical trials, clearing the way for new therapies.

Although the technology is not yet approved for routine clinical use, it’s constantly improving and aims to address the limitations inherent in today’s pathology processes.

“You do a biopsy, but instead of having a pathologist read it with their very rough scores of stage 1, 2, or 3, you read it by an AI-driven machine that can quantify it with a score of 1.5 or 1.75 instead of 1 or 2,” Vlad Ratziu, MD, PhD, professor of hepatology at the Sorbonne Université and Hôpital Pitié-Salpêtrière Medical School in Paris, France, and coeditor of The Journal of Hepatology, said in an interview.

“The technology is automated, more sensitive to change, and more highly quantitative. It has implications for liver disease diagnoses, clinical trials, and treatments,” added Dr. Ratziu, who has written about the promise and challenges inherent in developing treatments for metabolic dysfunction–associated steatotic liver disease (MASLD).

To explore the potential impact of AI-powered technologies for the clinic, this news organization spoke with representatives from three companies identified by Dr. Ratziu as leaders in the field: HistoIndexPathAI, and PharmaNest. Each company uses proprietary technology augmented by AI, and their tools have been used in published trials.
 

Moving Toward Better Diagnoses and Disease Management

The traditional approach for staging liver fibrosis relies on trained pathologists manually evaluating stained tissue samples obtained from biopsies of the liver.

But this method, though still considered the gold standard, doesn’t always provide the granularity needed for an accurate diagnosis or a reliable assessment in clinical trials, said Dean Tai, PhD, HistoIndex’s cofounder and chief scientific officer.

Although noninvasive tests (NITs), alone or with traditional histologic examination, are increasingly used during clinical management because they are less invasive and more repeatable for disease monitoring, they are limited in their precision and ability to provide comprehensive information, Dr. Tai said. That’s because “no single NIT or single-dimensional measurement of a biomarker offers a full assessment of disease activity, fibrogenic drive, and fibrosis load.”

In contrast, AI provides “a highly reproducible and objective assessment of liver fibrosis severity,” he said. “It eliminates the variability associated with staining methods, while revealing changes in the nano-architecture and morphology of collagen fibers not discernible by the human eye or current NITs, especially in the early stages of fibrosis or in cases of simultaneous progression and regression.”

Mathieu Petitjean, PhD, founder and CEO of PharmaNest, has a similar view. 

Although degree of liver fibrosis is associated with long-term outcomes of patients with MASLD, “poor detection thresholds due to their categorical nature mean that small and relevant changes are not reflected by changes in staging,” he said. “The reliable detection [with AI] of subtle changes in the phenotypes of fibrosis will significantly enrich the understanding of progression and regression of fibrosis severity.”

The ability of AI-based tools to see patterns the human eye cannot also means they could “help in predicting which patient may respond to a drug, in order to get the right treatments to the right patients as soon as possible,” said Katy Wack, PhD, vice president of clinical development at PathAI.

“Additionally, AI-based algorithms have been developed to provide more quantitative continuous scores to better capture change and discover new tissue-based biomarkers, which may be prognostic or predictive of clinical benefit,” she said. 

Such tools are currently undergoing testing and validation for use in trials and diagnostically.

The standardization and reproducibility offered by AI-driven technology could facilitate more consistent diagnoses across different healthcare settings, Dr. Tai suggested. “As the integration of the technology with other blood-, imaging-, and omics-based techniques evolves, it may enable earlier detection of liver diseases, more accurate monitoring of disease progression, and better evaluation of treatment responses, ultimately improving patient care and outcomes.”
 

 

 

More Effective Clinical Trials

The limitations of conventional pathology may be responsible, at least in part, for the repeated failure of novel compounds to move from phase 2 to phase 3 clinical trials, and from clinical trials to approval, the sources agreed.

“In clinical trials, patients are subject to enrollment criteria using liver biopsies, which are scored with a composite scoring system involving four different histologic components to grade and stage the disease,” Dr. Wack noted. 

However, there is wide variability between pathologists on biopsy scoring, and an individual pathologist presented with the same sample may give it a different score after some time has passed, she said.

That means “we are using a nonstandardized and inconsistent scoring system to determine whether a patient can be enrolled or not into a trial,” Dr. Wack said. 

The change in the composite score over a follow-up period, usually 1-2 years, determines whether a patient has responded to the candidate drug and, ultimately, whether that drug could be considered for approval, Dr. Wack said.

Because scores at the baseline and follow-up timepoints are not precise and inconsistent across pathologist readers, and even the same reader over time, there are often many “false-positive” and “false-negative” responses that can result in potential therapeutics either failing or succeeding in clinical trials, she said.

To address this variability in biopsy scoring as it relates to clinical trials, regulatory bodies have recommended a consensus approach, in which multiple pathologists read the same biopsy independently and a median score is used, or pathologists convene to come to an agreement, Dr. Wack said. 

“This is a very costly and burdensome approach and is still subject to interconsensus panel variation,” she said.

The introduction of digital pathology using validated digital viewers, where pathologists can view a glass slide digitally and pan and zoom over the image as they can with a microscope, means that many pathologists can read the same slide in parallel, she explained.

“If they need to discuss, they can do so efficiently over a phone call, each using their own computer screen and shared annotation tools to facilitate their discussion.”

Although this consensus approach can improve consistency, it still leads to variability in scoring across different groups of pathologists, Dr. Wack said.

This is where AI-assisted pathology comes into play.

“With this approach, a pathologist still views the image digitally, but an algorithm has predicted and highlighted key features and recommended quantitative scores,” she said.

This approach has been shown to increase precision for pathologists, thereby increasing reproducibility and standardizing scoring across timepoints and clinical trials.
 

What’s Ahead

These AI tools could address pathology’s lack of scalability, the result of a limited number of trained pathologists capable of doing liver biopsy assessments, Dr. Tai said. 

“Digital pathology workflows enable the transformation of conventional histologic glass slides into large digital images using scanners, allowing significant productivity gains in terms of workflow and collaboration,” he said.

Although AI-assisted pathology tools are still being validated, their promise for improving diagnoses and uncovering new treatments is clear, the interviewees agreed.

Extending its use to stage fibrosis in other liver diseases, such as primary biliary cholangitis, primary sclerosing cholangitis, and alcoholic liver disease, is also in progress on an experimental basis but will take time to validate.

“The landscape will evolve quickly in the coming 3-4 years,” Dr. Petitjean predicted. “To start, their intended use will likely be limited to a decision-support tool to enhance the performance of pathologists and perhaps stratify or triage cases sent for routine vs expert review.”

Dr. Petitjean even suggested that the increasing role of NITs and the amount of data being generated prospectively and retrospectively around liver biomarkers could mean that liver biopsies might not be needed one day.

A version of this article appeared on Medscape.com.

Publications
Publications
Topics
Article Type
Sections
Disallow All Ads
Content Gating
No Gating (article Unlocked/Free)
Alternative CME
Disqus Comments
Default
Use ProPublica
Hide sidebar & use full width
render the right sidebar.
Conference Recap Checkbox
Not Conference Recap
Clinical Edge
Display the Slideshow in this Article
Medscape Article
Display survey writer
Reuters content
Disable Inline Native ads
WebMD Article