65th ASH Annual Meeting
Theme 5: Artificial intelligence
Application of AI in hematology has the potential to utilize datasets on a much larger scale than would be possible with traditional analytical methods. AI was the subject of a highly informative special interest session titled “AI in Hematology: Where Do You Stand in 2023?” It also featured in many presentations at ASH 2023, particularly regarding risk stratification to aid in treatment selection and diagnosis based on analysis of digital images, a few of which are highlighted below. These exciting approaches could expedite clinical development, improve diagnosis, and facilitate personalization of therapy. However, AI is still at a fledgling stage of development and these tools must be further honed to ensure accuracy and patient confidentiality.
Risk Stratification
A machine‑learning model predicted outcomes of allogeneic-HSCT in patients with AML, which could provide valuable input for clinical decision‑making alongside traditional risk stratification tools.1 A separate presentation reported that a deep-learning Graph Neural Network was used with longitudinal EHR data to predict survival in AML. This included evaluation of labs and histology that are not considered in the traditional risk stratification tools for allogeneic‑HSCT.2
Cancer Vaccine Design
Machine learning was also used to identify endogenous retroviral elements and design simulated personalized cancer vaccines for seven patients with hematologic malignancies, encompassing AML, ALL, MM, DLBCL, and CLL.3 The machine‑learning approach may result in treatment regimens for these diseases with greater treatment effect and more durable response.
Diagnosis
Several presentations explored ways in which AI can aid in diagnosis of hematologic malignancies. One presentation reported high accuracy (92.3%) of an AI algorithm in differentiating prefibrotic primary myelofibrosis from essential thrombocytopenia, which was trained on 32,226 patient-derived digitized whole slide images from diagnostic bone marrow biopsies.4 Another presentation reported on a novel stain-free diagnostic flow‑cytometric approach called “ghost cytometry,” which uses machine vision‑based characterization of bone marrow cells to rapidly diagnose acute leukemias.5 These new approaches could offer rapid, cost-effective, and accurate diagnosis of hematologic malignancies in clinical practice in the future.
AI, artificial intelligence; ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; ASH, American Society of Hematology; CLL, chronic lymphocytic leukemia; DLBCL, diffuse large B-cell lymphoma; EHR, electronic health record; HSCT, hematopoietic stem cell transplantation; MM, multiple myeloma.
1. Sánchez AH et al. ASH Annual Meeting 2023. Abstract 62; 2. Sinha R et al. ASH Annual Meeting 2023. Abstract 960; 3. Thygesen CB et al. ASH Annual Meeting 2023. Abstract 243; 4. Srisuwananukorn A et al. ASH Annual Meeting 2023. Abstract 901; 5. Kawamura Y et al. ASH Annual Meeting 2023. Abstract 906.