April 6, 2023
Read in 2 minutes
Source/Disclosure
Issuer:
Disclosure:
L’Imperio does not report relevant financial disclosures. See research for relevant financial disclosures of all other authors.
Important points:
- Histological features derived from machine learning may provide prognostic value when learned and scored by pathologists.
- This represents an “AI milestone,” but research is needed to validate reproducible scoring.
A prognostic study validated the use of artificial intelligence-derived tumor adipose features, according to the study results. This may aid in colon cancer risk stratification and provide prognostic value when integrated into practice.
“Prognostic markers for colorectal cancer are of great clinical interest. may help with treatment regimen and duration.” Vincenzo LimperioMD, An assistant professor of pathology at the University of Milano-Bicocca and a colleague wrote: JAMA network open“In this context, it has recently been demonstrated that the use of digital pathology tools can provide prognostic information regarding colon cancer using routine histopathological slides.

“This prognostic study represents a milestone for AI in pathology and medicine, demonstrating both the feasibility and prognostic potential of pathologist-based integration of features identified by machine learning. ” Vincenzo LimperioMD, wrote a colleague.
Image: Adobe Stock
“This allows us to identify tumor adipose features (TAFs), moderately to poorly differentiated tumor cells in close proximity to adipocytes, as machine learning-derived features that show promising independent prognostic value in stage II and III colorectal cancer cases. ) has been identified.”
L’Imperio et al. used data from 258 histopathological cases of colon adenocarcinoma (53% male, median age 67 years, stage II, n = 119, stage III, n = 139). As pathologist scoring of previously machine-identified histopathological features, learning correlated with survival.
Two pathologists blinded to patient outcome identified TAF in 47% of cases, with multiple lesions in 12% and extensive lesions in 24%. Pathologist agreement was 72% for all TAF scores and 90% for broad TAF compared with other classifications.
The researchers reported a “significant prognostic value” of pathologist-identified TAF using a binary threshold for overall survival (HR = 1.55; 95% CI, 1.07-2.25). , did not report on CRC disease-specific survival (HR = 1.86; 95% CI, 0.95). -3.62). However, there was a dose-dependent association with broad TAF and overall survival (HR = 1.87; 95% CI, 1.23-2.85) and disease-specific survival (HR = 2.29; 95% CI, 1.09-4.7). bottom.
In multivariate analysis, age (HR = 1.07; 95% CI, 1.05-1.09), disease stage (HR = 1.6; 95% CI, 1.03-2.51), extensive TAF (HR = 1.79; 95% CI, 1.14- 2.81) was retained. Disease stage (HR = 3.57; 95% CI, 1.39-9.18) and extensive TAF (HR = 2.19; 95% CI, 1.01-4.75) were independent prognostic factors for disease-specific survival.
“This prognostic study represents a milestone for AI in pathology and medicine, demonstrating both the feasibility and prognostic potential of pathologist-based integration of features identified by machine learning.” concluded L’Imperio and colleagues. “After demonstrating generalizable prognostic value and consistent scoring strategies among pathologists, AI-derived prognostic features may be used alongside well-established features in future cases. , allowing for further validation and clinical integration.”