Evaluation of the Usefulness of Machine Learning and Artificial Intelligence on Hyperspectral Images in the Diagnosis of Myelodysplastic Syndrome.

Hyperspectral imaging (HSI) has gained increasing use in pathological diagnosis in recent years. HSI captures spectral information at wavelengths beyond the visible range. We analyzed specimens from 36 cases diagnosed with myelodysplastic syndrome (MDS) based on bone marrow biopsy or clot specimens at Showa Medical University Fujigaoka Hospital and classified them into three groups based on dysplastic lineage. HSIs of the biopsy specimens were acquired using a pushbroom hyperspectral camera. Spectral data were extracted from annotated target cells and used to train machine learning classifiers. Pixel-level data were divided into training and evaluation sets at the sample level and cross-validated. Our HSI-based artificial intelligence system achieved high accuracy (up to 97%) in bone marrow pathology, outperforming previous RGB-based studies and demonstrating the feasibility of HSI for MDS diagnosis. To the best of our knowledge, this study is the first such evaluation on bone marrow specimens.