Automated digital image analysis and manual counting of Ki-67 proliferation index in patients with breast cancer: a comparative study

    loading  Checking for direct PDF access through Ovid

Abstract

Background

Immunohistochemical Ki-67 labeling index (LI) assessment is critical for treatment decision making, for defining prognostic subgroups, for and predicting treatment outcomes in patients with breast cancer (BC). Manual counting of 1000 tumor cells is routinely done to evaluate Ki-67 LI; however, this method is tedious, labor intensive, and significantly influenced by interobserver and intraobserver reproducibility.

Aim

To measure Ki-67 proliferation index objectively in a group of patients with BC using automated digital image analysis (DIA) method and to evaluate its consistency with manual counting method.

Patients and methods

A series of 89 consecutive patients with invasive BC was included. All patients were diagnosed by preoperative core needle biopsies at Department of Pathology of National Cancer Institute. Immunohistochemical analysis of estrogen receptor, progesterone receptor, HER2, and Ki-67 levels was done for all patients as a routine service. Ki-67 immunostained slides were assessed by visual assessment (VA) method and DIA method using Leica QWin image processing and analysis workstation. The scores of the hot spots were calculated. All cases were classified into three groups (<14, 14–30, and >30% Ki-67 LI). The concordance between VA and automated DIA for Ki-67 was evaluated, and the correlation of Ki-67 scores with histopathological parameters was determined.

Results

The intraclass correlation coefficients analysis showed substantial agreement between VA and DIA of Ki-67 in the whole group of the current study (intraclass correlation coefficients=0.768, 95% confidence interval: 0.668–0.842, P<0.001). VA and DIA Ki-67 index results agreed in 67.4% (60/89) of cases (K=0.295, P<0.001). The agreement between VA and DIA within the three Ki-67 index groups was highest (3/4; 75%) in Ki-67 index less than 14%, less (51/72; 70%) in Ki-67 LI more than 30%, and lowest (6/13; 46.2%) in the intermediate Ki-67 index 14–30%.

Conclusion

The current study showed a very good agreement between automated Ki-67 LI and human visual counts in BC cases. The use of an automated analyzer may reduce subjectivity and improve the accuracy and the reproducibility of Ki-67 LI assessment in BC cases specifically in the range of 10–30%.

Related Topics

    loading  Loading Related Articles