Lakshmi Harinath

Morphologic Evaluation of Hyperchromatic Crowded Groups Present in Cervical Thin Prep Cytology Tests as Identified by the Hologic Genius Digital Diagnostics System: A Descriptive Study

ABSTRACT Background Our descriptive study focused on morphologic characteristics of hyperchromatic crowded groups (HCGs) in ThinPrep cervical cytology tests when reviewed with the artificial intelligence (AI)‐assisted Hologic Genius Digital Diagnostics System (HGDDS). Method After IRB approval, our archives were searched over a 1‐year period for potential HCGs. A total of 157 slides with HCGs were selected, scanned, and analyzed using the HGDDS. One cytologist and one cytopathologist interpreted these cases while enumerating the cytomorphologic characteristics as seen with the HGDDS. Results Of the 157 cases, a total of 84.7% were called Negative for Intraepithelial Lesion or Malignancy (NILM) on original ThinPrep interpretation (OTPI) as opposed to 76.4% with HGDDS. 5.7% and 3.2% of the cases were called high‐grade squamous intraepithelial lesion (HSIL) and atypical glandular cells (AGC) on OTPI, as compared to 4.5% and 8.3% on HGDDS. 6.4% of cases were interpreted as adenocarcinoma on both OTPI and HGDDS. A total of 16 cases were called NILM‐Atrophy by both modalities. Concordance for pathologist diagnosis between HGDDS and OTPI for 157 cases was 0.610 (kappa value). For 25 cases, there was a follow‐up biopsy diagnosis, including 10 cases of adenocarcinoma, 5 of Cervical Intraepithelial Neoplasia (CIN) 2–3, and 1 case of CIN 1. The sensitivity, specificity, positive predictive value, and negative predictive value for the detection of CIN2+ lesions, when ASC‐H/AGC and above were considered, were 100%, 50%, 75%, and 100%, respectively. Conclusion Our initial study shows encouraging results in the evaluation of HCGs presented as two‐dimensional static images on a computer monitor by the HGDDS.

Artificial Intelligence in Gynecologic Cytology

Background: Cervical cancer is the fourth most common cancer in women globally with highest incidence and mortality identified in less developed and medically underserved areas in the world. The diminishing cytology workforce, unavailability of expert consultation, and the high volume of Pap tests needing manual screening are the main reasons for exploring innovative solutions to help mitigate the negative effects resulting from the dearth of timely cervical cancer screening in certain population groups. Summary: Developments in whole slide imaging and artificial intelligence (AI) have enabled the emergence of new computer-assisted systems that have the potential for transforming traditional cytopathology practice. However, AI-based systems are relatively new with limited published data on their validation and clinical utility in clinical practice. Our article aims to increase awareness of the availability of such systems, explores the history and development of AI-assisted screening platforms for Pap tests, compares the performance characteristics of various systems, elaborates on technical challenges associated with conducting clinical trials employing this technology, and discusses considerations around deploying such systems in routine cytopathology practice. Key Message: Revolutionary AI-based systems are being developed and utilized in cytopathology practice to screen Pap tests. Some of these systems have good performance characteristics and provide opportunities to combat various issues such as workload and standardization faced by cytology laboratories globally. However, judicious review of these systems using evidence-based studies is imperative to promote widespread adoption and maintain high-quality standards for patient safety.

Assessment of reprocessed ThinPrep cytology cases after glacial acetic acid wash procedure using the Hologic Genius Digital Diagnostics System

This study focuses on ThinPrep Pap tests with a low to borderline number of cells and the performance of AI-assisted digital systems in cases that have undergone the acetic acid wash procedure (AAW). Four hundred sixty-two cases initially interpreted as unsatisfactory and finally interpreted as satisfactory after AAW procedure were included in the study. These ThinPrep Pap slides were scanned using the Genius Digital Diagnostic System (GDDS). Overall agreement between GDDS and Original ThinPrep Interpretation (OTPI) was 63.2% for diagnostic match (Negative for Intraepithelial Lesion, ASCUS, Low Grade Squamous Intraepithelial Lesion, Atypical Squamous Cells, High Grade, Atypical Glandular Cells, or unsatisfactory), and 66.0% when ASCUS + diagnoses are grouped. Out of the 462 cases, 364 (78.8%) were called Negative for Intraepithelial Lesion based upon the manual OTPI, as opposed to 310 (67.1%) reviewed using the GDDS. There were 17.5%, 1.3%, 0.9% and 1.5% cases called Atypical Squamous Cells of Undetermined Significance, Low Grade Squamous Intraepithelial Lesion, Atypical Squamous Cells, High Grade Cannot be Excluded, and Atypical Glandular Cells respectively on OTPI, as opposed to 24.7%, 3.0%, 0.6% and 1.5% respectively by the GDDS. Only 3.0% of the cases were deemed unsatisfactory by GDDS. All the cases with high grade results in the subsequent cervical biopsy were diagnosed as at least Atypical Squamous Cells of Undetermined Significance and above by the GDDS. The diagnostic agreement between GDDS and biopsy was 65.2% compared to 58.7% for OTPI, although this is not statistically significantly different, (χ Our results demonstrate that the GDDS can be successfully used to screen ThinPrep Pap Tests that have undergone the AAW procedure.

Assessment of the efficacy and accuracy of cervical cytology screening with the Hologic Genius Digital Diagnostics System

Abstract Background Medical technologies powered by artificial intelligence are quickly transforming into practical solutions by rapidly leveraging massive amounts of data processed via deep learning algorithms. There is a necessity to validate these innovative tools when integrated into clinical practice. Methods This study evaluated the performance of the Hologic Genius Digital Diagnostics System (HGDDS) with a cohort of 890 previously reviewed and diagnosed ThinPrep Papanicolaou (Pap) tests with the intent to deploy this system for routine clinical use. The study included all diagnostic categories of The Bethesda System, with follow‐up tissue sampling performed within 6 months of abnormal Pap test results to serve as the ground truth. Results The HGDDS demonstrated excellent performance in detecting significant Pap test findings, with close to 100% sensitivity (98.2%–100%) for cases classified as atypical squamous cells of undetermined significance and above within a 95% confidence interval and a high negative predictive value (92.4%–100%). Conclusions The HGDDS streamlined workflow, reduced manual workload, and functioned as a stand‐alone system.

Analysis of the sensitivity of high‐grade squamous intraepithelial lesion Pap diagnosis and interobserver variability with the Hologic Genius Digital Diagnostics System

AbstractBackgroundArtificial intelligence (AI)–based systems are transforming cytopathology practice. The aim of this study was to evaluate the sensitivity of high‐grade squamous intraepithelial lesion (HSIL) Papanicolaou (Pap) diagnosis assisted by the Hologic Genius Digital Diagnostics System (GDDS).MethodsA validation study was performed with 890 ThinPrep Pap tests with the GDDS independently. From this set, a subset of 183 cases originally interpreted as HSIL confirmed histologically were included in this study. The sensitivity for detecting HSIL by three cytopathologists was calculated.ResultsMost HSIL cases were classified as atypical glandular cell/atypical squamous cell–high grade not excluded (AGC/ASC‐H) and above by all cytopathologists. Of these cases, 11.5% were classified as low‐grade squamous intraepithelial lesion (LSIL) by pathologist A (P‐A), 6% by pathologist B (P‐B), and 5.5% by pathologist C (P‐C); 3.8%, 2.7%, and 1.6% of these cases were classified as atypical squamous cell of unknown significance (ASC‐US) by P‐A, P‐B, and P‐C, respectively. The sensitivity for detection of cervical intraepithelial neoplasia 2 and above (CIN2+) lesions was 100% if ASC‐US and above (ASC‐US+) abnormalities were counted among all three pathologists. The sensitivity for detection of CIN2+ lesions was 84.7%, 91.3%, and 92.9% by P‐A, P‐B, and P‐C, respectively, for ASC‐H and above abnormalities. The Kendall W coefficient was 0.722, which indicated strong agreement between all pathologists.ConclusionsNew‐generation AI‐assisted Pap test screening systems such as the GDDS have the potential to transform cytology practice. In this study, the GDDS aided in interpreting HSIL in ThinPrep Pap tests, with good sensitivity and agreement between the pathologists who interacted with this system.

The clinical significance of atypical glandular cells in Papanicolaou tests: changes in diagnostic patterns over 15 years at a single institution

Detection of atypical glandular cells (AGCs) by Papanicolaou (Pap) test remains a significant challenge in gynecological cytology. We compared follow-up diagnoses, age groups, and human papillomavirus (HPV) results for AGC at our institution to that of our previous study (study period 2008-2013). AGC Paps diagnosed and HPV results between January 2020 and June 2024 were obtained from the database at UPMC Magee-Womens Hospital. Of the total 188,320 Paps performed during the study period, 1025 had AGC diagnoses comprising 0.54% of the total. A total of 92.2% of cases had a companion HPV test, with positive HPV results seen in 32.9% of cases. Overall, 33.3% (286/859) of AGC cases had subsequent significant histologic findings (cervical intraepithelial neoplasia 2 and 3, adenocarcinoma in-situ, endocervical adenocarcinoma, endometrial lesions, metastatic carcinoma). Detection of cervical lesions was highest in women <30 years (50%) and significantly decreased with increasing age (P < 0.0001). Identification of endometrial lesions was highest in the ≥50-year group (P < 0.0001). Nearly half of AGC/HPV-positive cases had significant cervical findings, while these were detected in only 2.1% of AGC/HPV-negative cases (P < 0.0001). Endometrial lesions were identified in 25.7% of AGC/HPV-negative cases, but only in <1% of AGC/HPV-positive cases (P < 0.0001). Significant differences were identified comparing the 2 study periods: increased HPV testing (P < 0.0001), increased HPV-positivity (P = 0.0029), decreased AGC rate (P < 0.0001), and increased endometrial lesions on follow-up (P < 0.0001). Our findings continue to support HPV results and patient age as valuable data in triaging AGC. AGC/HPV-positive results frequently suggest a cervical/HPV-related lesion, often in younger patients. Conversely, AGC/HPV-negative results, especially in patients ≥50 years, support noncervical lesional origins.

Diagnostic performance of the hologic genius digital diagnostics system for low-grade squamous intraepithelial lesion (LSIL) ThinPrep papanicolaou tests

Advancements in digital imaging technology for Papanicolaou test slides, combined with artificial intelligence are driving the development and adoption of innovative computer-assisted screening methods for cervical cancer within the cytology community. Our study aimed to assess the performance of the Hologic Genius Digital Diagnostic System (HGDDS) in the interpretation of low-grade squamous intraepithelial lesions (LSIL) in ThinPrep Papanicolaou slides. As part of a validation study performed with 890 ThinPrep Papanicolaou slides using the HGDDS, a subset of 146 LSIL cases were included in this study. Performance characteristics for the detection of cervical intraepithelial neoplasia (CIN) and interobserver variability among 3 cytopathologists were assessed. On evaluation of the consensus results of the 3 cytopathologists, of the 146 LSIL Papanicolaou cases, 60.3% were interpreted as LSIL with the HGDDS. The remainder were interpreted as ASCUS (26%), ASC-H (10.3%), HSIL (2.7%), and NILM (0.7%). The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for detecting CIN1+ lesions in the ASCUS + category with the HGDDS were 100%, 25%, 97.9%, and 100%, respectively. The sensitivity, specificity, PPV, and NPV for the detection of CIN1+ lesions in the LSIL + category with the HGDDS were 74.7%, 75%, 99.1%, and 7.7%, respectively. Kendall's W coefficient was 0.792, indicating strong agreement among participating pathologists. Our study demonstrated that ThinPrep Papanicolaou tests with LSIL could be interpreted with strong agreement among pathologists and with good performance indicators when utilizing the HGDDS.