Artificial intelligence could improve the detection of binuclear erythroblasts (BNE), a rare and difficult-to-quantify phenomenon that can indicate myelodysplastic syndromes (MDS), according to a new report. That's it.
The researchers behind the study say this new approach streamlines the use of new technology and makes leveraging machine learning more feasible. This research scientific report.1
The authors explained that MDS is particularly heterogeneous but can usually be diagnosed on the basis of morphological bone marrow (BM) dysplasia and persistent cytopenias.
Myelodysplastic syndromes (MDS) are a group of bone marrow diseases. | Image credit: © Sviatlana – Stock.adobe.com
“However, accurate diagnosis of cases with mild cytopenias or subtle dysplastic changes can be difficult, and inter-scorer variability and subjectivity can exist even among experienced hematopathologists. “Yes,” the researchers wrote.
Some patients are left with uncertain diagnoses, such as idiopathic cytopenia of undetermined significance (ICUS) or clonal cytopenia of undetermined significance (CCUS).
Given the current lack of precision, the researchers said it is important to identify an objective, standardized method to differentiate MDS from non-clonal reactive causes of cytopenias and dysplasia. .
“Furthermore, while rare phenomena indicative of MDS, such as binuclear erythroblasts, are easily identified using visual microscopy, when they occur in large numbers they can be difficult to quantify, making statistical robustness difficult. may be limited,” the researchers said.
One possible solution is imaging flow cytometry (IFC). “This is because it combines the high-throughput data collection capabilities and statistical robustness of traditional multicolor flow cytometry (MFC) with the high-resolution imaging capabilities of a microscope in one system.”
Previous studies by the same group found IFC to be effective in analyzing morphometric changes in erythropoietic-deficient BM cells.2 In that study, researchers used IFC to analyze samples from 14 MDS patients, 6 ICUS/CCUS patients, 6 non-MDS controls, and 11 healthy controls.
The researchers found that the IFC model “reliably identified true binucleate erythroblasts at a significantly higher frequency at two of the three stages of erythroblast maturation in MDS patients compared to normal BM. and enumerated.” P = .0001).
Still, they state that the workflow for feature-based IFC analysis is difficult, time-consuming, and requires software-specific expertise. Therefore, in a new paper, he proposed to analyze IFC image data using a convolutional neural network (CNN) algorithm. They state that the CNN algorithm has higher accuracy and more flexibility in data interpretation than feature-based analysis alone. Additionally, we used specially designed artificial intelligence software with a graphical user interface designed to display meaningful results for researchers without advanced coding skills.
To test the new method, the researchers used raw data from a previous study, analyzed it using a new artificial intelligence model, and compared the model's results to previous IFC analysis. Each sample was also manually examined to verify the presence of BNE.
The accuracy of the new model was 94.3% and the specificity was 98.2%. The latter means that the model is unlikely to misclassify non-BNE and BNE. The sensitivity of the model was low as 21.1% of his BNEs in the dataset were incorrectly classified as erythroblasts exhibiting irregular nuclear morphology. But overall, the researchers said, the data suggests high confidence that when the model identifies BNE, it's correct.
The researchers said it was notable that the model performed as well despite the smaller dataset they used to train it. They stated that model performance is likely to improve by incorporating more robust datasets.
“Emphasis should be placed on enriching classes of cells with irregular nuclear morphology and BNE that cause classification difficulties,” the researchers wrote. “Furthermore, it may be beneficial to expand the scope of classification categories to include categories of uncertain cases in addition to BNE, doublets, and cells with irregular nuclear morphology.”
But for now, researchers said their studies show that AI has the potential to be an effective and efficient diagnostic tool for MDS patients.
References:
- Rosenberg CA, Rodriguez MA, Bill M, Ludvigsen M. Comparative analysis of feature-based ML and her CNN for binuclear erythroblast quantification in myelodysplastic syndrome patients using imaging flow cytometry data. science officer. 2024;14(1):9349. doi:10.1038/s41598-024-59875-x
- Rosenberg CA, Bill M, Rodriguez MA, et al. Imaging flow cytometry and machine learning-assisted morphometry to explore abnormalities in erythropoiesis in patients with myelodysplastic syndromes. Cytometry B Clincytome. 2021;100(5):554-567. doi:10.1002/cyto.b.21975
