Myelodysplastic syndromes (MDS) are a group of hemopoietic clonal disorders characterized by peripheral blood cytopenias and a cellular bone marrow (BM) showing cell atypias that reflect abnormalities in proliferation, maturation and apoptosis of hemopoietic precursors 1 2 3 4 5 6 . According to WHO criteria 7 8 9 the percentage of blasts counted in BM cytology is an essential parameter for diagnosis and classification of the several types of MDS, as well as for the differential diagnosis between refractory anemia with excess of blasts (RAEB) and acute myeloid leukemia. In normal hemopoiesis, strict morphological criteria can easily be used to define each stage of cell maturation, but in MDS, immature cells presenting an asynchronous maturation may often be difficult to classify 4 6 7 8 9 . Standardized morphologic criteria have been recommended 8 9 in order to separate MDS blasts and MDS promyelocytes. Yet, there is always some degree of subjectivity, although the FAB Group 1 and the International Working Group on MDS 8 9 had defined classification criteria, which were also included in the 2008 WHO classification 7 . Moreover, the European LeukemiaNet published in the Internet a consensus-based cell library elaborated by experienced morphologists 8 9 that could be used as a guide for daily work and training. Despite of these efforts, the morphologic diagnosis continues to be a difficult task, and the morphologic diagnosis of MDS should only be done by a consensus of two expert morphologists 3 6 .

In cytological bone marrow smears of MDS patients, some immature cells may not be classifiable in a satisfactory way, because they show simultaneously characteristics of blasts and promyelocytes, thus not fulfilling the criteria of either category. This problem is known to the practicing hematologist, but, surprisingly not discussed in the scientific literature.

In recent years, virtual microscopy and computerized image analysis gained increasing importance 10 11 12 13 14 . These techniques have been widely used in pathology and cytology for the differentiation of normal cells, benign and malignant tumors 15 16 17 , as prognostic markers in malignancy 18 19 20 and in order to examine chromatin remodeling of cells in culture after incubation with carcinogens 21 , hormones 22 and therapeutic agents 23 24 . Computerized image analysis has shown to be a fast and reliable way for quantitative morphologic analysis 10 18 19 25 26 27 28 , and moreover, to be a possibility to detect subtle morphologic changes which cannot be recognized by conventional microscopy even by an expert.

The aim of the present study was to examine whether computerized nuclear texture analysis could help to characterize in a more objective way the blasts and promyelocytes in normal bone marrow, as well as in patients with MDS. We also wanted to examine if this technology was able to classify atypical immature myeloid cells present in bone marrow smears of these patients.

Diagnostic smears from 19 normal BM and 30 cases of MDS were examined. Median age of the patients with normal BM was 57 years (24 - 85; 9 males and 10 females). The median age of the MDS patients was 64 years (30 - 85). There were 16 males and 14 females (Table 1). According to the WHO type, most cases were RCMD (refractory cytopenia with multilineage dysplasia) and RAEB (refractory anemia with excess of blasts).

In normal BM, nuclei of promyelocytes showed significant changes, such as an increase of perimeter and local texture homogeneity and a decrease in form factor, chromatin gray levels, Haralick's entropy, inertia, energy, contrast, diagonal moment, cluster prominence, Minkowski's fractal dimension and its goodness-of-fit when compared with that of normal blasts (Table 2).

Compared to normal myeloblast nuclei, chromatin texture of MDS blasts had a higher local homogeneity as well as goodness-of-fit of the fractal dimension, but a lower entropy, contrast, diagonal moment, and fractal dimension (Table 3). Nuclei of promyelocytes from MDS showed a larger nuclear area, local homogeneity and goodness-of-fit of FD, but lower inertia, entropy, energy, contrast, diagonal moment and FD than normal ones (Table 4).

The values obtained for nuclei of granulocytic precursors in MDS are presented in Table 5.

Nuclei of atypical myeloid precursors showed intermediate characteristics between those of blasts and promyelocytes according to the quantitative features perimeter, form factor, gray level and its standard deviation, but were similar to promyelocytes according to the texture variables inertia, energy, contrast, diagonal moment, cluster prominence and Minkowski's fractal dimension. None of the features studied showed closeness to myeloblasts. These findings were independent of the WHO type of MDS.

In recent years, computerized image analysis has been widely used in histology and cytology in order to examine tissue differentiation, for tumor classification 11 12 15 16 17 and for the search of new prognostic variables in neoplasias 18 19 20 . This approach has also been used in basic research for analyzing nuclear texture changes that reflect chromatin remodeling of cells after incubation with carcinogens 21 , hormones 22 and therapeutic agents 23 24 .

Chromatin remodeling, which is primarily due to epigenetic events, can be found during cell differentiation or malignant transformation 20 21 23 24 26 . In normal hemopoiesis, gene expression during normal cell maturation is controlled by genetic and epigenetic changes 30 . Blast cells in acute myeloid leukemia and MDS always present epigenetic abnormalities and their DNA methylation signature is different to that of any stage of normal myeloid maturation, an observation which permits to distinguish normal and leukemic blasts 30 31 32 33 .

In the present study we examined the utility of computerized chromatin texture analysis for the diagnosis of normal and atypical immature myeloid precursors in routine BM smears. In normal BM, all except three of the quantitative features examined, presented significant differences between blasts and promyelocytes. Thus, chromatin texture analysis in routine BM cytology is able to define cells in early stages of myeloid maturation.

In MDS, genetic and epigenetic alterations provoke abnormalities of proliferation, maturation, and apoptosis 5 6 32 33 34 , which, of course, are reflected in subtle alterations of the chromatin structure. The expression of various lineage and maturation-related membrane proteins may be discordant in granulopoietic precursors 32 33 34 provoking also morphological atypias. All these alterations hamper the cell classification. The difficulties in classify several cases with MDS are well known. The European LeukemiaNet created a consensus-based cell library elaborated by experienced morphologists and downloaded it in the Internet 9 . Its purpose is to be a guide for daily work and training. Furthermore, it is still recommended that the morphologic diagnosis of MDS should be achieved by a consensus of two experienced morphologists 3 6 . Our study underlines these problems of classification. The cells diagnosed as atypical immature myeloid precursors in the present study did not - on light microscopic examination - reveal a Golgi apparatus, and would be therefore morphologically classified as blasts. However, the nucleus showed features more close to that of promyelocytes or nuclei of an "intermediate" state between blasts and promyelocytes.

Computerized texture analysis confirmed this subjective impression, since 4 features presented intermediate values between blasts and promyelocytes and 6 variables pointed out a similarity with nuclei of promyelocytes. Although the cytoplasm in these atypical cells still reveals characteristics of a blast, the nuclear structure is more similar to a promyelocyte, thus indicating an asynchronous maturation in MDS patients.

The existence of this intermediate maturation stage is of clinical importance. Cytologists basing their diagnosis mainly on cytoplasmic criteria will count these atypical immature cells together with blasts and thus increase the blast count. Other observers, emphasizing the similarity of the nuclear features, might count them together with promyelocytes, or count them separately, thus diminishing the blast count. This may imply in a different classification for the patient.

The BM blast count is considered very important for the classification of MDS in the revised WHO classification 7 as well as for the determination of the IPSS and WPSS scores 2 31 32 . Furthermore, the blast count is considered to be an independent prognostic feature of utmost importance in MDS 2 6 7 33 35 36 37 . Therefore, additional pathophysiological and molecular studies should be performed in order to investigate whether these atypical precursor cells should not be counted as blasts, as suggested by our investigation.

The authors declare that they have no competing interests.

JRV: collected all the clinical data, selected the material, classified the cells, acquired and segmented the cell images and reviewed the literature. RLA: developed the segmentation software and made all the computational analysis. ILM: provided the conception and design of the study, microscopic analysis, classification of the cells, drafting of the manuscript, and final approval. KM: participated in the study design, performed the statistical analyses and interpretation of the data, and made the critical revision of the text.

This study is part of a master thesis of JRV (Postgraduate Course in Medical Pathophysiology, FCM, University of Campinas) with ILM as principal advisor and KM as co-advisor. All authors have read and approved the manuscript.