Abstract
Early diagnosis and treatment initiation of chronic myeloid leukemia (CML) are considered to increase the rate of deep molecular response. However, the early diagnosis of CML is challenging due to the absence of clinical symptoms and peripheral blood test anomaly, especially at the timing of peripheral white blood cell count is within a normal range. This study explored the possibility of artificial intelligence (AI)-based quantitative detection of CML cells using ghost cytometry (GC) technology. We created pre-trained AI models, using the morphological information data of the peripheral blood leukocytes obtained from patients newly diagnosed with CML and healthy individuals. We applied these models to peripheral blood samples from CML patients after initiating tyrosine kinase inhibitor treatment at various time points, which contains smaller amounts of tumor cells. The AI model accurately detected CML cells and a strong correlation between AI-detected CML cells and actual BCR::ABL1(IS) mRNA levels was observed. We concluded that the multidimensional morphological information of single cells obtained using GC, combined with machine learning algorithms, enables the quantitative detection of label-free CML cells. Our finding may enable the development of a screening test that can identify early-stage patients with CML before numerical abnormalities appear in blood tests.