Image Analysis Detects Lineage-Specific Morphologic Markers in Leukemic Blast Cells

Abstract

This report outlines the morphologic classification of acute myeloid (AML: French-American-British FAB classification: Ml) and lymphoid (ALL) leukemia by automatic image analysis and the correlation to immunologic and cytochemical classification. The investigation was carried out on Romanowsky-Giemsa stained bone marrow (n = 15) and blood smears (n -10) from 25 patients with primary acute leukemia. The cases had been classified as of myeloid or lymphoid origin by three hcmatologic centers using imtnunochemistry or cytochemistry, but the specimens were submitted to the authors' laboratory without the diagnosis. The nuclear and cytoplasmic pattern of the blast cells were analyzed by a high resolution image analysis system and the measured and The FAB classification has established a widely accepted morphologic categorization of acute leukemias with improved reproducibility and standardization.1 Major problems in applying morphologic classifications arise from the subjective impression of the observer and thus only a 60% to 80% degree of reproducibility. 23 However, morphologic cell classification based on high resolution computer aided image analysis is independent from subjective impression. 4 Therefore, the resulting cell analysis is based on objective and reproducible measurable criteria. This is primarily done by morphology, immunophenotype, and cytochemistry and additionally by cytogenetics and molecular genetics.The subtypes of acute myeloid leukemias (AML) are mainly diagnosed by morphologic and cytochemical features according to the FAB classification. 1710 Immunophenotyping appears to be indispensable in distinguishing the AML subtypes M0 and Ml with minimal signs of cellular maturation from acute lymphoid leukemia (ALL). Apart from the identification of promyelocytic leukemia FAB M3, neither morphologic nor immunologic classification of AML proved to be of prognostic relevance.10 However, the subtypes of ALL are distinguished predominantly by means of their immunophenotype, which proved to be of prognostic relevance, 6 " whereas morphologic assessment plays a minor role, apart from the subtypes with FAB L2 morphology. 1213In this study, we present a morphologic characterization of AML (FAB: Ml) and ALL by image analysis and 23 Downloaded from https://academic.oup.

Probabilistic Model‐Based Cell Tracking

“…Furthermore, with imaging data ever more simply and rapidly acquired, manual tracking becomes progressively impractical. As a result, automated cell tracking systems are mandatory to further advance the study of biological cells [2,3,6,8,[18][19][20].…”

“…The cell detection problem is essentially one of anomaly detection: the localization of groups of pixels inconsistent with the random behavior of the image background. A wide variety of semi automatic or automatic methods have been proposed to segment cell boundaries [2,3] which can be divided into three major categories.…”

“…(2) We empirically define a size range r ∈ [2,4] pixels, within which the cell radius is uniformly distributed. (3) Any hypothesis which has cells violating the minimum required separation between cells is assigned a probability of zero.…”

“…The corresponding automation of the analysis of such huge sets of images would allow fundamentally new questions to be addressed in proteomics, genomics, and stem-cell research [1][2][3][4][5][6][7][8]. This paper proposes a coupling of advanced methods in pattern recognition and image processing [9][10][11][12][13][14][15][16] to an existing cell-imaging platform [17] in which the analysis, currently being undertaken by hand, is impossibly slow and tedious for the volumes of data being generated.…”

Constrained watershed method to infer morphology of mammalian cells in microscopic images

“…Manual tracking of such data is an onerous task, and yet still widely practiced, so automated methods are in high demand. Cell segmentation in microscopic images as an object segmentation problem remains an attractive and challenging task due to the often corrupted or blurred images, high noise, the presence of clutter, and the difficulties of adapting and extending available image segmentation approaches (10–13).…”