Session: 634. Myeloproliferative Syndromes: Clinical and Epidemiological: Poster II
Hematology Disease Topics & Pathways:
Artificial intelligence (AI), MPN, Chronic Myeloid Malignancies, Diseases, Myeloid Malignancies, Technology and Procedures, Imaging, Machine learning, Pathology
Myeloproliferative neoplasms (MPNs) are a group of clonal hematopoietic stem cell disorders characterized by the excessive proliferation of 1 or more myeloid cell lineages. The evaluation of bone marrow pathology is essential for diagnosing and classifying MPNs. However, morphological assessments of bone marrow trephine (BMT) section by hematopathologists are inherently subjective. Therefore, we aim to develop a potential auxiliary artificial-based diagnostic tool to assist hematopathologists in evaluating bone marrow metrics accurately and objectively.
Methods:
We collected hematoxylin-eosin (H&E) and Gomori staining trephine sections and clinical information from 282 patients with MPNs (78 ET patients, 37 pre-PMF patients, 167 PMF patients) and 33 nonneoplastic patients (25 iron-deficiency anemia (IDA) patients and 8 healthy donors) in a single center.
Based on U2-Net, UNeXt, and ResNet, we developed an automatic quantitative analyzing platform for bone marrow trephine (BMT) sections for patients with MPNs in evaluating bone marrow cellularity, the myeloid-to-erythroid (M: E) ratio, megakaryocyte morphology and distribution, and the severity of marrow fibrosis. A bone marrow classification model incorporating bone marrow metrics, a clinical classification model utilizing clinical features, and a comprehensive classification model that includes both bone marrow metrics and clinical features were developed using random forest classifiers to differentiate MPN subtypes and non-MPN.
Results:
Bone marrow metrics, including marrow cellularity, the myeloid-to-erythroid (M: E) ratio, megakaryocyte morphology and distribution, and the severity of marrow fibrosis (MF) were precisely quantitatively analyzed (with an accuracy of approximately 0.9) based on the accuracy segmentation and identification of various cells and tissues (with an intersection over union (IoU) of approximately 0.8).
The bone marrow and clinical classification models reached a micro-average area under the curve (AUC) of 0.94 for differentiating non-MPN and MPN subtypes. Compared to the clinical classification model, the bone marrow classification model performed better in most cases, other than in distinguishing ET and other MPN or non-MPN (AUC 0.86 vs. 0.92). The comprehensive classification model showed higher accuracy on the whole, with a micro-average AUC of 0.98. The AUC was 0.94 for discriminating between ET and other samples, 0.84 for discriminating between pre-PMF and other samples, 0.99 for discriminating between PMF and other samples, and 1 for discriminating between non-MPN and MPN samples.
Conclusion:
This platform is highly accurate for quantitatively analyzing bone marrow pathology and classifying MPN subtypes and non-MPN. It can be a potentially auxiliary diagnostic tool for hematopathologists when dealing with patients with MPNs.
Disclosures: Zhang: Takeda (China) International Trading Co., Ltd.: Consultancy, Honoraria, Research Funding.
See more of: Oral and Poster Abstracts