Article Type
Changed
Mon, 07/13/2020 - 14:29

Artificial intelligence (AI)–based Gleason scores correlated with pathologists’ Gleason scores for predicting survival in patients with prostate cancer, according to a cohort-based analysis.

An AI-based Gleason score – derived from 7,267 digitized biopsy slides, pathology reports, and clinical data from patient electronic medical records – was calculated for each of 599 prostate cancer patients.

The AI scores were compared with pathologists’ Gleason scores, which were obtained from pathology reports for each of the patients.

The two scores were “highly correlated,” according to investigators. The area under the curve (AUC) for the 7-year mortality rate was 0.667 for the AI-based scores and 0.659 for the pathologists’ scores.

The investigators also found that markers extracted using AI-based algorithms could predict disease progression in patients with low- and higher-grade disease.

Daphna Laifenfeld, PhD, chief scientific officer of Ibex Medical Analytics in Tel Aviv, reported these results in a poster at the AACR virtual meeting II. Ibex Medical Analytics is the company that developed the AI-based algorithms and Gleason score (the Ibex score).

In addition to comparing the Ibex Gleason scores with pathologists’ scores, Dr. Laifenfeld and colleagues sought to “develop AI markers – computational features extracted from slides using AI-based algorithms – that can predict disease progression in low-, and separately, higher-grade patients.”

Information extracted using the algorithms included Gleason scores; perineural invasion; and other characteristics such as inflammation, high-grade prostatic intraepithelial neoplasia, and atrophy.

“We used data ... to address each aim, analyzing hundreds of patients in each comparison, and employed logistic regression to develop the predictive models,” Dr. Laifenfeld said.

Of the 357 patients evaluated, 180 had low-grade disease, defined by a prebiopsy prostate-specific antigen (PSA) level less than 10 ng/mL (Gleason group 1), and 177 patients had higher-grade disease (Gleason group 2 or higher).

Gleason group 1 patients were considered to have progressed if they developed higher-grade cancer, underwent prostatectomy, or if their cancer had metastasized. Gleason group 2 and above patients were considered to have progressed if their cancer metastasized or if they had a postprostatectomy PSA level greater than 4 ng/ml.

In Gleason group 1 patients, combining multiple features from the pathology report with prebiopsy PSA levels was shown to predict disease progression better than prebiopsy PSA levels alone (AUC, 0.687).

“Importantly, AI markers that combine features automatically extracted by Ibex with prebiopsy PSA levels are even better associated with progression (AUC, 0.748),” Dr. Laifenfeld said.

Similarly, in the Gleason group 2 and above patients, the AI markers that combine Ibex-extracted features with prebiopsy PSA levels were also highly associated with progression (AUC, 0.862 vs. AUC, 0.77 for the non–Ibex-based approach) and can be used for patient stratification, Dr. Laifenfeld said.

“For each patient, we can predict whether or not their disease will progress,” she said. “[T]his type of stratification can then be used to support clinical disease management decisions, and [it can be used] in the course of drug development for patient stratification and trial enrichment strategies.”

Dr. Laifenfeld and some coinvestigators are employed by Ibex Medical Analytics.

SOURCE: Laifenfeld D et al. AACR 2020, Abstract 867.

Meeting/Event
Publications
Topics
Sections
Meeting/Event
Meeting/Event

Artificial intelligence (AI)–based Gleason scores correlated with pathologists’ Gleason scores for predicting survival in patients with prostate cancer, according to a cohort-based analysis.

An AI-based Gleason score – derived from 7,267 digitized biopsy slides, pathology reports, and clinical data from patient electronic medical records – was calculated for each of 599 prostate cancer patients.

The AI scores were compared with pathologists’ Gleason scores, which were obtained from pathology reports for each of the patients.

The two scores were “highly correlated,” according to investigators. The area under the curve (AUC) for the 7-year mortality rate was 0.667 for the AI-based scores and 0.659 for the pathologists’ scores.

The investigators also found that markers extracted using AI-based algorithms could predict disease progression in patients with low- and higher-grade disease.

Daphna Laifenfeld, PhD, chief scientific officer of Ibex Medical Analytics in Tel Aviv, reported these results in a poster at the AACR virtual meeting II. Ibex Medical Analytics is the company that developed the AI-based algorithms and Gleason score (the Ibex score).

In addition to comparing the Ibex Gleason scores with pathologists’ scores, Dr. Laifenfeld and colleagues sought to “develop AI markers – computational features extracted from slides using AI-based algorithms – that can predict disease progression in low-, and separately, higher-grade patients.”

Information extracted using the algorithms included Gleason scores; perineural invasion; and other characteristics such as inflammation, high-grade prostatic intraepithelial neoplasia, and atrophy.

“We used data ... to address each aim, analyzing hundreds of patients in each comparison, and employed logistic regression to develop the predictive models,” Dr. Laifenfeld said.

Of the 357 patients evaluated, 180 had low-grade disease, defined by a prebiopsy prostate-specific antigen (PSA) level less than 10 ng/mL (Gleason group 1), and 177 patients had higher-grade disease (Gleason group 2 or higher).

Gleason group 1 patients were considered to have progressed if they developed higher-grade cancer, underwent prostatectomy, or if their cancer had metastasized. Gleason group 2 and above patients were considered to have progressed if their cancer metastasized or if they had a postprostatectomy PSA level greater than 4 ng/ml.

In Gleason group 1 patients, combining multiple features from the pathology report with prebiopsy PSA levels was shown to predict disease progression better than prebiopsy PSA levels alone (AUC, 0.687).

“Importantly, AI markers that combine features automatically extracted by Ibex with prebiopsy PSA levels are even better associated with progression (AUC, 0.748),” Dr. Laifenfeld said.

Similarly, in the Gleason group 2 and above patients, the AI markers that combine Ibex-extracted features with prebiopsy PSA levels were also highly associated with progression (AUC, 0.862 vs. AUC, 0.77 for the non–Ibex-based approach) and can be used for patient stratification, Dr. Laifenfeld said.

“For each patient, we can predict whether or not their disease will progress,” she said. “[T]his type of stratification can then be used to support clinical disease management decisions, and [it can be used] in the course of drug development for patient stratification and trial enrichment strategies.”

Dr. Laifenfeld and some coinvestigators are employed by Ibex Medical Analytics.

SOURCE: Laifenfeld D et al. AACR 2020, Abstract 867.

Artificial intelligence (AI)–based Gleason scores correlated with pathologists’ Gleason scores for predicting survival in patients with prostate cancer, according to a cohort-based analysis.

An AI-based Gleason score – derived from 7,267 digitized biopsy slides, pathology reports, and clinical data from patient electronic medical records – was calculated for each of 599 prostate cancer patients.

The AI scores were compared with pathologists’ Gleason scores, which were obtained from pathology reports for each of the patients.

The two scores were “highly correlated,” according to investigators. The area under the curve (AUC) for the 7-year mortality rate was 0.667 for the AI-based scores and 0.659 for the pathologists’ scores.

The investigators also found that markers extracted using AI-based algorithms could predict disease progression in patients with low- and higher-grade disease.

Daphna Laifenfeld, PhD, chief scientific officer of Ibex Medical Analytics in Tel Aviv, reported these results in a poster at the AACR virtual meeting II. Ibex Medical Analytics is the company that developed the AI-based algorithms and Gleason score (the Ibex score).

In addition to comparing the Ibex Gleason scores with pathologists’ scores, Dr. Laifenfeld and colleagues sought to “develop AI markers – computational features extracted from slides using AI-based algorithms – that can predict disease progression in low-, and separately, higher-grade patients.”

Information extracted using the algorithms included Gleason scores; perineural invasion; and other characteristics such as inflammation, high-grade prostatic intraepithelial neoplasia, and atrophy.

“We used data ... to address each aim, analyzing hundreds of patients in each comparison, and employed logistic regression to develop the predictive models,” Dr. Laifenfeld said.

Of the 357 patients evaluated, 180 had low-grade disease, defined by a prebiopsy prostate-specific antigen (PSA) level less than 10 ng/mL (Gleason group 1), and 177 patients had higher-grade disease (Gleason group 2 or higher).

Gleason group 1 patients were considered to have progressed if they developed higher-grade cancer, underwent prostatectomy, or if their cancer had metastasized. Gleason group 2 and above patients were considered to have progressed if their cancer metastasized or if they had a postprostatectomy PSA level greater than 4 ng/ml.

In Gleason group 1 patients, combining multiple features from the pathology report with prebiopsy PSA levels was shown to predict disease progression better than prebiopsy PSA levels alone (AUC, 0.687).

“Importantly, AI markers that combine features automatically extracted by Ibex with prebiopsy PSA levels are even better associated with progression (AUC, 0.748),” Dr. Laifenfeld said.

Similarly, in the Gleason group 2 and above patients, the AI markers that combine Ibex-extracted features with prebiopsy PSA levels were also highly associated with progression (AUC, 0.862 vs. AUC, 0.77 for the non–Ibex-based approach) and can be used for patient stratification, Dr. Laifenfeld said.

“For each patient, we can predict whether or not their disease will progress,” she said. “[T]his type of stratification can then be used to support clinical disease management decisions, and [it can be used] in the course of drug development for patient stratification and trial enrichment strategies.”

Dr. Laifenfeld and some coinvestigators are employed by Ibex Medical Analytics.

SOURCE: Laifenfeld D et al. AACR 2020, Abstract 867.

Publications
Publications
Topics
Article Type
Click for Credit Status
Ready
Sections
Article Source

FROM AACR 2020

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