Self-Supervised Learning Can Distinguish Myelodysplastic Neoplasms from Clinical Mimics Using Bone Marrow Biopsies

The diagnosis of myelodysplastic neoplasms (MDS) requires examination of the bone marrow for morphologic evidence of dysplasia. We sought to determine if a self-supervised learning (SSL) AI image analysis approach may be utilized to reliably distinguish MDS from its clinically relevant mimics using bone marrow biopsies (BMBx). Whole slide images (WSIs) of H&E- and reticulin-stained BMBx sections from 243 unique patients (89 MDS, 55 non-MDS cytopenic controls [NMCC], and 99 negative control [NC] cases) were segmented into tiles and analyzed. These tiles were then processed using the Barlow Twins SSL model to generate histomorphologic phenotype clusters (HPCs). Review of the HPCs revealed the clusters enriched in MDS captured known histopathologic features of MDS including hypercellularity, dysplastic and clustered megakaryocytes, increased immature hematopoietic cells, increased vascularity, fibrosis, and cell streaming patterns. Assessment of 95 MDS BMBx images from a second institution showed consistent HPC enrichment patterns, validating the robustness of the model. The trained ensemble model using H&E- and reticulin-stained slides distinguished MDS from NCs with an AUC of 0.89, and from age-matched, NMCCs with an AUC of 0.84. These findings demonstrate the potential of SSL approaches to capture diagnostically relevant morphologic patterns and to improve the reproducibility of MDS diagnosis.