Presented During:
Fri, 11/8: 4:45 PM - 5:00 PM
Hyatt Regency Orlando
Submission No:
1251
Submission Type:
Platform or Poster
First Author:
Lakshmi Harinath, MD MPH
University of Pittsburgh/UPMC
Co-Author(s):
Esther Elishaev, MD
University of Pittsburgh Medical center
Yuhong Ye, MD, PhD
The first affiliated Hospital of Fujian University, China
Jonee Matsko, SCT, MB(ASCP)
Magee Womens Hospital of UPMC
Amy Colaizzi, SCT
University of Pittsburgh Medical center
Stephanie Wharton, BS, SCT (ASCP)
N/A
Liron Pantanowitz, MD, PhD, MHA
UPMC/Pitt
Chengquan Zhao, MD
Magee Womens Hospital of UPMC
Introduction:
Revolutionary artificial intelligence (AI)-based systems are transforming cytopathology practice. The Hologic Genius Digital Diagnostic System (GDDS) combines AI algorithmic analysis with volumetric imaging technology in the interpretation of ThinPrep Pap slides. Our study determines the sensitivity of HSIL interpretation by the GDDS and evaluates interobserver variability amongst cytopathologists.
Materials and Methods:
A validation study was performed with 890 ThinPrep Paps where a cytotechnologist and three cytopathologists reviewed all of the cases on GDDS independently. A total of 183 cases originally interpreted as HSIL with histologic diagnosis of CIN2+ lesions confirmed by biopsy were included in this study. Sensitivity for detecting HSIL by three cytopathologists was calculated. Kendall's W coefficient was applied to calculate consistency of their interpretations.
Results:
Most of the HSIL cases were classified as AGC/ASC-H and above by all cytopathologists. 11.5% of the cases were classified as LSIL by Pathologist A (PA), 6% by Pathologist B (PB) and 5.5 % by Pathologist C (PC). 3.8%, 2.7% and 1.6% cases were classified as ASCUS by PA, PB and PC respectively (Table 1). The sensitivity for detection of CIN2+ lesion was 100% if ASCUS and above lesions were counted amongst all three pathologists. The sensitivity for detection of CIN2+ lesions was 84.7%, 91.3%, and 92.9% by PA, PB and PC respectively, for ASC-H and above lesions (Table 2). Kendall's W coefficient was 0.722, indicating strong agreement between all pathologists.
Conclusions:
AI-assisted Pap screening has the potential to further change cytology practice. GDDS aids in interpreting HSIL in ThinPrep Pap tests with high sensitivity and strong agreement between pathologists who interacted with this system. The variation in classification could be due to non-availability of clinical history, HPV status and learning involved during diagnosis. Additional studies focusing on workflow, AI-assisted diagnoses for all categories of The Bethesda System and outcomes are needed as more labs adopt the GDDS.
Presentation Category:
Emerging Tools and Technology (Includes Digital Cytology, AI + Information Technology)
Awards:
All accepted abstracts will be considered for the Geno Saccomanno, MD Award and New Frontiers in Cytology Award.
To be considered for other awards, please select below all that apply.
No Awards