The counting and classification of blood cells allow for the evaluation and diagnosis of avast number of diseases. The analysis of white blood cells (WBCs) allows for the detection of acute lym-phoblastic leukaemia (ALL), a blood cancer that can be fatal if left untreated. Currently, the morphologicalanalysis of blood cells is performed manually by skilled operators. However, this method has numerousdrawbacks, such as slow analysis, non-standard accuracy, and dependences on the operator’s skill. Fewexamples of automated systems that can analyse and classify blood cells have been reported in the liter-ature, and most of these systems are only partially developed. This paper presents a complete and fullyautomated method for WBC identification and classification using microscopic images.Methods: In contrast to other approaches that identify the nuclei first, which are more prominent thanother components, the proposed approach isolates the whole leucocyte and then separates the nucleusand cytoplasm. This approach is necessary to analyse each cell component in detail. From each cellcomponent, different features, such as shape, colour and texture, are extracted using a new approach forbackground pixel removal. This feature set was used to train different classification models in order todetermine which one is most suitable for the detection of leukaemia.Results: Using our method, 245 of 267 total leucocytes were properly identified (92% accuracy) from 33images taken with the same camera and under the same lighting conditions. Performing this evaluationusing different classification models allowed us to establish that the support vector machine with aGaussian radial basis kernel is the most suitable model for the identification of ALL, with an accuracyof 93% and a sensitivity of 98%. Furthermore, we evaluated the goodness of our new feature set, whichdisplayed better performance with each evaluated classification model.Conclusions: The proposed method permits the analysis of blood cells automatically via image processingtechniques, and it represents a medical tool to avoid the numerous drawbacks associated with manualobservation. This process could also be used for counting, as it provides excellent performance and allowsfor early diagnostic suspicion, which can then be confirmed by a haematologist through specialisedtechniques.

Leucocyte classification for leukaemia detection using image processing techniques

PUTZU, LORENZO;CAOCCI, GIOVANNI;DI RUBERTO, CECILIA
2014-01-01

Abstract

The counting and classification of blood cells allow for the evaluation and diagnosis of avast number of diseases. The analysis of white blood cells (WBCs) allows for the detection of acute lym-phoblastic leukaemia (ALL), a blood cancer that can be fatal if left untreated. Currently, the morphologicalanalysis of blood cells is performed manually by skilled operators. However, this method has numerousdrawbacks, such as slow analysis, non-standard accuracy, and dependences on the operator’s skill. Fewexamples of automated systems that can analyse and classify blood cells have been reported in the liter-ature, and most of these systems are only partially developed. This paper presents a complete and fullyautomated method for WBC identification and classification using microscopic images.Methods: In contrast to other approaches that identify the nuclei first, which are more prominent thanother components, the proposed approach isolates the whole leucocyte and then separates the nucleusand cytoplasm. This approach is necessary to analyse each cell component in detail. From each cellcomponent, different features, such as shape, colour and texture, are extracted using a new approach forbackground pixel removal. This feature set was used to train different classification models in order todetermine which one is most suitable for the detection of leukaemia.Results: Using our method, 245 of 267 total leucocytes were properly identified (92% accuracy) from 33images taken with the same camera and under the same lighting conditions. Performing this evaluationusing different classification models allowed us to establish that the support vector machine with aGaussian radial basis kernel is the most suitable model for the identification of ALL, with an accuracyof 93% and a sensitivity of 98%. Furthermore, we evaluated the goodness of our new feature set, whichdisplayed better performance with each evaluated classification model.Conclusions: The proposed method permits the analysis of blood cells automatically via image processingtechniques, and it represents a medical tool to avoid the numerous drawbacks associated with manualobservation. This process could also be used for counting, as it provides excellent performance and allowsfor early diagnostic suspicion, which can then be confirmed by a haematologist through specialisedtechniques.
Image processing, Microscopic image segmentation, Cell analysis, White blood cell detection, Leukaemia classification
File in questo prodotto:
File Dimensione Formato  
AIM2014_open.pdf

accesso aperto

Descrizione: Articolo principale
Tipologia: versione post-print
Dimensione 3.28 MB
Formato Adobe PDF
3.28 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11584/94592
Citazioni
  • ???jsp.display-item.citation.pmc??? 26
  • Scopus 195
  • ???jsp.display-item.citation.isi??? 139
social impact