New method of nuclear grading of tissue sections by means of digital image analysis with prognostic significance for node‐negative breast cancer patients

• Published in: Cytometry (New York, N.Y.) Vol. 24; no. 2; pp. 140 - 150
• Main Authors: Albert, R., Müller, J. G., Kristen, P., Schindewolf, T., Kneitz, S., Harms, H.
• Format: Journal Article
• Language: English
• Published: New York Wiley‐Liss, Inc 01.06.1996
• Subjects: Algorithms; Breast Neoplasms - pathology; Cell Nucleus - physiology; chromatin structure analysis; Computer‐aided nuclear grading; digital image processing; Female; Follow-Up Studies; Humans; Microtomy; prognostic factors; Recurrence; Retrospective Studies; Signal Processing, Computer-Assisted
• ISSN: 0196-4763; 1097-0320; 1097-0320
• DOI: 10.1002/(SICI)1097-0320(19960601)24:2<140::AID-CYTO6>3.0.CO;2-N

To optimize treatment of the individual patient with node‐negative breast cancer, objective, reproducible, and standardized prognostic criteria are required. A number of factors have been studied in recent years, but until now it has been possible to obtain information about the risk of recurrence only for some patients belonging to subgroups with special characteristics. We report the establishment of an image analysis method for nuclear grading as an attempt to solve this problem. In a retrospective analysis, we used routine hematoxylin and eosin‐stained paraffin sections from 54 node‐negative patients with surgery between 1980 and 1985. Cell scenes of primary tumors were scanned in a light microscope in successive focus positions to obtain three‐dimensional information. After automatic image segmentation, nuclear features were calculated as input for a first binary classification tree to differentiate between tumor and nontumor cells. Tumor nuclei from patients with or without relapse were defined as high‐risk or low‐risk nuclei, respectively, and were separated with a second tree. Feature values of the measured tumor nuclei from each patient were examined with this second tree to analyze whether the majority of nuclei for each patient were classified as high‐risk or low‐risk nuclei. Correct classification rates in the two binary cell classification trees were 88.0% and 83.8%, respectively. In the learning sample of our study, all patients with a relapse had the majority of nuclei in the high‐risk group, most with more than 80%. Therefore, it seems to be possible to develop an image analytical risk profile system for nuclear grading to provide information on individual prognosis. © 1996 Wiley‐Liss, Inc.