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

Paper Details

Abstracts, authors, affiliations, and disclosures.

Monday, December 14, 2026
06:00 p.m. - 08:00 p.m.
MCCNO - Hall F
Technology and Procedures, Assays, Imaging, Molecular Testing, Newborn Screening, Omics Technologies, Pathology, Serologic Tests, Bioinformatics, Computational Biology, Emerging Technologies, Profiling, Treatment Considerations, Combination Therapy, Therapy Sequence, Biological Therapies, Antibody Therapy, Transplantation, Bispecific Antibody Therapy, (CAR)-T Cell Therapies, Monoclonal Antibody Therapy, Non-Biological Therapies, Chemotherapy, Surgical, Adverse Events, Artificial Intelligence (AI), Computer Vision, Deep Learning, Generative AI, Machine Learning, Natural Language Processing, Biological Processes, Pathogenesis, CHIP, Genomics, Hematopoiesis, Immune Mechanism, Metabolism, Molecular Biology, Multi-Systemic Interactions, Clinical Practice (Health Services and Quality), Clinical Guidelines, Diseases, Hemoglobinopathies, Sickle Cell Disease, Immunology, GVHD, Lymphoid Malignancies, Lymphoid Leukemias, Lymphomas, Plasma Cell Disorders, Myeloid Malignancies, Acute Myeloid Malignancies, Chronic Myeloid Malignancies, Measurable Residual Disease, ALL, CLL, Non-Hodgkin Lymphoma, B Cell Lymphoma, Indolent Lymphoma, Aggressive Lymphoma, Research, Fundamental Science, AML, APL, Translational Research, MDS, MPN, CMML, Clinical Research, Clinical Trials, Health Outcomes Research, Patient-Reported Outcomes, Real-World Evidence, Registries, Survivorship, Study Population, Human, Adult, Elderly, Pediatric
PAPER

6101 Artificial intelligence-based assessment of bone marrow slides to distinguish response and residual disease after induction in acute myeloid leukemia (AML)

Utsav Joshi, MD, Palak Dave, PhD2, Sameer Andani, MD3, Joseph Pierce4, Najla Al Ali, MS5, David Sallman, MD6, Rami Komrokji, MD7, Jeffrey Lancet, MD8, Issam El Naqa, PhD9, Ling Zhang, MD10, Onyee Chan, MD11
1 Moffitt Cancer Center, Tampa, Florida, United States
2 Moffitt Cancer Center, Tampa, Florida, United States
3 Moffitt Cancer Center, Tampa, Florida, United States
4 Moffitt Cancer Center, Tampa, Florida, United States
5 Moffitt Cancer Center, Tampa, Florida, United States
6 Moffitt Cancer Center, Tampa, Florida, United States
7 Moffitt Cancer Center, Tampa, Florida, United States
8 Moffitt Cancer Center, Tampa, Florida, United States
9 Moffitt Cancer Center, Tampa, Florida, United States
10 Moffitt Cancer Center, Tampa, Florida, United States
11 Moffitt Cancer Center, Tampa, Florida, United States

Introduction: Short-term response to induction therapy in AML is typically assessed by bone marrow (BM) specimen on day 14 (D14) or day 21 (D21). Marked hypoplasia (<10–20% cellularity) with <5% residual blasts on D14/D21 BM slides is highly predictive of complete remission by day 28. However, patients (pts) with >5% blasts are a heterogeneous group with some representing true residual disease, while others could still be on course to remission. Expert pathology review of BM slides may not distinguish the subtle differences between these two, leading to over-treatment with second induction and unnecessary toxicity. Artificial Intelligence (AI)-based examination of BM slides could provide a reproducible and standardized tool to resolve this diagnostic ambiguity.

Methods: We evaluated adults with non-acute promyelocytic leukemia AML treated with frontline intensive induction, focusing on borderline D14/D21 BM samples (5-30% blasts, ≤15% cellularity) digitized at 40× magnification. Ground truth classification (residual AML vs delayed morphologic remission) was established using follow-up BM assessments documented in pts charts. Pts who underwent second induction due to immediate residual disease were excluded to prevent response interpretation bias. Model performance was evaluated using 10-repeated 3-fold cross-validation on a cohort of 66 pts (training cohort, N=44; validation cohort, N=22 per fold). To differentiate these trajectories, we utilized our previously reported AI-based BM evaluation framework- a BM whole-slide imaging evaluation pipeline that integrates three complementary components to analyze bone marrow biopsy (BMB), bone marrow aspirate smears (BMA), and BMA-derived Bag-of-Cells (BoC). Tile-level features from BMA and BMB whole-slide images were extracted using the H-Optimus-0 pathology foundation model and aggregated via attention-based multiple instance learning (ABMIL) to obtain slide level representation. To capture fine-grained cell morphology, a BoC representation was derived from the BMA by extracting single-cell features using the DinoBloom-B foundation model followed by ABMIL. Late-fusion strategies (weighted averaging, logistic regression, XGBoost, calibrated logistic regression, multilayer perceptron) and feature-level fusion strategies (concat fusion and gated attention fusion) were evaluated to assess the value of integrating these complementary representations.

Results: This cohort included 66 pts with evaluable post-induction data. The most common induction regimens were CPX-351 (N=22, 33.3%) and 7+3-based combination (N=35, 53%). Median BM blast percentage after induction was 9% (IQR 6-14.3%), and median marrow cellularity after induction was 5.8% (IQR 5-8%). Eighteen pts (27.3%) had a post-induction blast percentage in the borderline 5-6% range. Among the three pts with a post-induction blast percentage of 30%, all had marrow cellularity ≤10%. Following induction, 31 pts (47%) had residual disease, and 35 pts (53%) achieved remission. Among the individual representations, the BoC representation achieved the strongest performance (AUROC 0.714±0.090), outperforming both BMA (0.641±0.115) and BMB (0.598±0.118). While predictions from the three modalities demonstrated low pairwise correlation (Pearson r = 0.02-0.24), suggesting distinct model behavior, prediction error analysis showed substantial overlap in failure patterns (error correlation r = 0.62-0.68). None of the evaluated fusion approaches outperformed the strongest individual modality (BoC; AUROC 0.714±0.090), with the best fusion strategy logistic regression (LR) achieving AUROC 0.708±0.083. Subgroup analysis by blast percentage showed that in pts with ≤10% blasts (N=42), the BoC model achieved an AUC of 0.658±0.134 with LR performance at 0.652±0.130, whereas in those with >10% blasts (N=24), model efficacy was higher, yielding the BoC model AUC of 0.777±0.194 and the LR performance of 0.807±0.195. 

Conclusion: This proof-of-concept study shows that AI models built on pathology foundation models can meaningfully distinguish residual AML from delayed remission using BMA and BMB slides, with BMA-derived BoC representations showing the most promise. While the performance of the BoC model and best fusion model was comparable for the subgroup with ≤10% blasts, the fusion model outperformed the BoC model on the subgroup with >10% blast. Future work will focus on validation in larger, multi-institutional cohorts.

Utsav Joshi, MD: No relevant conflicts of interest to declare.