M Beil1, T Irinopoulou, J Vassy, J P Rigaut. 1. Laboratory of Image Analysis in Cell Pathology, Hospital Saint Louis, University of Paris, France.
Abstract
OBJECTIVE: Three-dimensional (3D) images of tissue blocs can be obtained through optical sectioning by a confocal scanning laser microscope and by piling up of consecutive confocal planes. These images offer the opportunity to quantitate biologic structures (i.e., nuclear chromatin) in 3D. Chromatin texture analysis is regarded as a key element in nuclear grading of malignant lesions. The present study introduced a new method of structural texture analysis of 3D chromatin distribution in cell nuclei. The method was assessed for its validity in a pilot study. STUDY DESIGN: The study group consisted of 300 3D cell nuclei from prostatic lesions, 100 each of hyperplasia, dysplasia and well-differentiated carcinoma. A set of 15 texture features was studied for its classification capability in a discriminant analysis. RESULTS: A classifier that confined three texture features yielded a classification accuracy of 91.7% for the training set (60 nuclei). The classification accuracy for the test set (240 nuclei) was 93.3%. CONCLUSION: The new method demonstrated its potential for 3D chromatin analysis. Further studies on large numbers of cases are necessary to evaluate the use of 3D nuclear grading for diagnostic pathology.
OBJECTIVE: Three-dimensional (3D) images of tissue blocs can be obtained through optical sectioning by a confocal scanning laser microscope and by piling up of consecutive confocal planes. These images offer the opportunity to quantitate biologic structures (i.e., nuclear chromatin) in 3D. Chromatin texture analysis is regarded as a key element in nuclear grading of malignant lesions. The present study introduced a new method of structural texture analysis of 3D chromatin distribution in cell nuclei. The method was assessed for its validity in a pilot study. STUDY DESIGN: The study group consisted of 300 3D cell nuclei from prostatic lesions, 100 each of hyperplasia, dysplasia and well-differentiated carcinoma. A set of 15 texture features was studied for its classification capability in a discriminant analysis. RESULTS: A classifier that confined three texture features yielded a classification accuracy of 91.7% for the training set (60 nuclei). The classification accuracy for the test set (240 nuclei) was 93.3%. CONCLUSION: The new method demonstrated its potential for 3D chromatin analysis. Further studies on large numbers of cases are necessary to evaluate the use of 3D nuclear grading for diagnostic pathology.