2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
05:30 p.m. - 07:30 p.m.
MCCNO - Hall F
Access, Representation, Population Health, Technology and Procedures, Assays, Molecular Testing, Omics Technologies, Treatment Considerations, Combination Therapy, Drug Development, Drug-Drug Interactions, Supportive Care, Therapy Sequence, Biological Therapies, Antibody Therapy, Bispecific Antibody Therapy, Immunotherapy, Non-Biological Therapies, Adverse Events, Artificial Intelligence (AI), Machine Learning, Biological Processes, Apoptosis, Genomics, Immune Mechanism, Clinical Practice (Health Services and Quality), Clinical Guidelines, Health Economics, Patient Safety, Practice Models, Quality of Care, Systems-Based Hematology, Diseases, Bleeding and Clotting, Infectious Diseases, Lymphoid Malignancies, Lymphoid Leukemias, Lymphomas, Measurable Residual Disease, CLL, B Cell Lymphoma, Aggressive Lymphoma, Research, Translational Research, Clinical Research, Clinical Trials, Epidemiology, Health Outcomes Research, Patient-Reported Outcomes, Real-World Evidence, Registries, Study Population, Human, Adult, Elderly
PAPER

2132 Artificial intelligence - assisted workflow for chronic lymphocytic leukemia (CLL) FISH – a real-world clinical experience

Xinjie Xu, PhD, FACMG, Michael Pitzer2, Reid Meyer3, Christopher A Sattler4, Kassandra Fisch5, Ben Black6, Falon Markow7, Sara Sprague, PhD8, Ben Cahoon9, Jansen N Seheult, MBBCh, BAO, MSc, MS, MD10
1 Mayo Clinic, Rochester, United States
2 Mayo Clinic, Rochester, United States
3 Mayo Clinic, Rochester, United States
4 Mayo Clinic, Rochester, United States
5 Mayo Clinic, Rochester, United States
6 Techcyte, Inc., Orem, United States
7 Techcyte, Inc., Orem, United States
8 Techcyte, Inc., Orem, United States
9 Techcyte, Inc., Orem, United States
10 Mayo Clinic, Rochester, United States

Background

Diagnostic evaluation of chronic lymphocytic leukemia (CLL) includes fluorescence in situ hybridization (FISH) testing for recurrent cytogenetic abnormalities involving chromosomes 6, 11, 12, 13, and 17, as well as IGH::CCND1 rearrangement. These findings provide important diagnostic and prognostic information. The Mayo Clinic FISH laboratory uses MetaSystems image scanners for automated acquisition of fluorescent probe signals in hematologic malignancy FISH testing. Previously, a computer vision model was developed and evaluated for the detection and classification of enumeration and fusion FISH signal patterns in multi–Z-stack darkfield microscopy field-of-view (FOV) images (1,2). Building upon this work, a semi-automated workflow was developed and validated on the Techcyte platform to incorporate AI-assisted analysis for the CLL FISH panel, including five enumeration probes (6q, 11q, chromosome 12, 13q, and 17p) and one fusion probe (IGH::CCND1) (3). The AI-assisted workflow was implemented clinically in April 2026.

Methods and Results

Approximately 1,750 CLL FISH cases with 10,000 FISH probe sites were analyzed using the AI-assisted workflow between May and October 2026. Ongoing post-implementation quality monitoring includes both object-level metrics (precision and recall) and scan-level metrics (reclassification rate and Jensen–Shannon [JS] divergence) for all six probes in the CLL panel. Reclassification rate serves as an overall performance metric reflecting the proportion of AI classifications requiring technologist correction and is a key predictor of workflow efficiency gains. JS divergence quantifies the similarity between AI-generated classifications and technologist-verified results.

Across the five enumeration probes, precision ranged from 84.18% to 91.31% and recall ranged from 88.05% to 93.56%. Reclassification rates ranged from 5.21% to 9.86%, while JS divergence ranged from 0.025 to 0.037. Performance of the fusion model for the IGH::CCND1 probe was lower, with a precision of 78.96% and recall of 76.0%. The IGH::CCND1 model demonstrated a reclassification rate of 18.97% and a JS divergence of 0.053. Overall, these metrics demonstrate robust performance of the enumeration models and acceptable performance of the fusion model in real-world clinical CLL cases. Six month average abnormal case rate is 63.23% using previous manual analysis workflow vs. 61.76% using AI assisted workflow, demonstrating equivalency between two workflows.

In addition to analytical performance, technologist review times were evaluated. Following implementation of the AI-assisted workflow, average technologist analysis time per case was reduced by approximately 50%. Despite these efficiency gains, 3%–10% of FISH sites still required manual microscopic review because of suboptimal image quality or ambiguous/complex signal patterns.

Conclusions

Our real-world clinical experience demonstrates that an AI-assisted FISH workflow can deliver accurate results while substantially improving analysis efficiency for CLL FISH testing. These findings support broader implementation of AI-enabled image analysis workflows across additional FISH assays using scanned digital images. Validation of AI-assisted workflows for AML and MDS FISH panels is currently underway, with planned clinical implementation in December 2026. Transition of the remaining hematologic FISH assays to AI-assisted workflows is anticipated in 2027.

References

Xu X. et al. Object detection model for quantifying fluorescence in situ hybridization (FISH) signal patterns in chronic lymphocytic leukemia (CLL). International Society for Laboratory Hematology (ISLH) Annual Meeting, Halifax, Nova Scotia, Canada. 2025 Xu X. et al. Object detection model for identification of fusion fluorescence in situ hybridization (FISH) probe signal patterns in hematologic malignancies. American Society of Human Genetics (ASHG) Annual Meeting, Boston, MA, USA 2025 Xu X. et al. Validation of an AI model assisted workflow for enumeration and fusion fluorescence in situ hybridization (FISH) probes in chronic lymphocytic leukemia (CLL). Association of Molecular Pathology (AMP) Annual Meeting, Boston, MA, USA 2025
Xinjie Xu, PhD, FACMG: No relevant conflicts of interest to declare.