Vanna R, Ronchi P, Lenferink A T M, Tresoldi C, Morasso C, Mehn D, Bedoni M, Picciolini S, Terstappen L W M M, Ciceri F, Otto C, Gramatica F
Laboratory of Nanomedicine and Clinical Biophotonics, Fond. Don Carlo Gnocchi ONLUS, Piazzale Morandi 6, 20121, Milan, Italy.
Analyst. 2015 Feb 21;140(4):1054-64. doi: 10.1039/c4an02127d.
In clinical practice, the diagnosis and classification of acute myeloid leukaemia (AML) and myelodysplastic syndrome (MDS) start from the manual examination of stained smears of bone marrow (BM) and peripheral blood (PB) by using an optical microscope. This step is subjective and scarcely reproducible. Therefore, the development of subjective and potentially automatable methods for the recognition of typical AML/MDS cells is necessary. Here we have used Raman spectroscopy for distinguishing myeloblasts, promyelocytes, abnormal promyelocytes and erhytroblasts, which have to be counted for a correct diagnosis and morphological classification of AML and MDS. BM samples from patients affected by four different AML subtypes, mostly characterized by the presence of the four subpopulations selected for this study, were analyzed. First, each cell was scanned by acquiring 4096 spectra, thus obtaining Raman images which demonstrate an accurate description of morphological features characteristic of each subpopulation. Raman imaging coupled with hierarchical cluster analysis permitted the automatic discrimination and localization of the nucleus, the cytoplasm, myeloperoxidase containing granules and haemoglobin. Second, the averaged Raman fingerprint of each cell was analysed by multivariate analysis (principal component analysis and linear discriminant analysis) in order to study the typical vibrational features of each subpopulation and also for the automatic recognition of cells. The leave-one-out cross validation of a Raman-based classification model demonstrated the correct classification of myeloblasts, promyelocytes (normal/abnormal) and erhytroblasts with an accuracy of 100%. Normal and abnormal promyelocytes were distinguished with 95% accuracy. The overall classification accuracy considering the four subpopulations was 98%. This proof-of-concept study shows that Raman micro-spectroscopy could be a valid approach for developing label-free, objective and automatic methods for the morphological classification and counting of cells from AML/MDS patients, in substitution of the manual examination of BM and PB stained smears.
在临床实践中，急性髓系白血病（AML）和骨髓增生异常综合征（MDS）的诊断与分类始于使用光学显微镜对骨髓（BM）和外周血（PB）染色涂片进行人工检查。这一步骤具有主观性且几乎无法重复。因此，开发用于识别典型AML/MDS细胞的主观且可能可自动化的方法很有必要。在此，我们使用拉曼光谱法区分原粒细胞、早幼粒细胞、异常早幼粒细胞和幼红细胞，这些细胞对于AML和MDS的正确诊断及形态学分类是必须计数的。对受四种不同AML亚型影响的患者的BM样本进行了分析，这些亚型大多以本研究选择的四个亚群的存在为特征。首先，通过采集4096个光谱对每个细胞进行扫描，从而获得拉曼图像，这些图像准确描述了每个亚群的形态特征。拉曼成像结合层次聚类分析能够自动区分和定位细胞核、细胞质、含髓过氧化物酶的颗粒和血红蛋白。其次，通过多变量分析（主成分分析和线性判别分析）对每个细胞的平均拉曼指纹进行分析，以研究每个亚群的典型振动特征，并用于细胞的自动识别。基于拉曼的分类模型的留一法交叉验证表明，原粒细胞、早幼粒细胞（正常/异常）和幼红细胞的正确分类准确率为100%。正常和异常早幼粒细胞的区分准确率为95%。考虑四个亚群的总体分类准确率为98%。这项概念验证研究表明，拉曼显微光谱法可能是一种有效的方法，可用于开发无标记、客观且自动的方法，用于对AML/MDS患者的细胞进行形态学分类和计数，以替代对BM和PB染色涂片的人工检查。
