ECR 2013 Rec: Texture analysis of malignant breast tumours: is a differentiation of ductal carcinoma in situ, invasive ductal and invasive lobular breast cancer possible? #B0680 #SS1302
B-0680 Texture analysis of malignant breast tumours: is a differentiation of ductal carcinoma in situ, invasive ductal and invasive lobular breast cancer possible?
T. Knogler, K. Pinker-Domenig, N. Perry, S. Milner, K. Mokbel, M.E. Mayerhoefer | Sunday, March 10, 10:30 – 12:00 / Room F2
Purpose: To evaluate the ability of texture features (TF), to differentiate between ductal carcinoma in situ (DCIS), invasive ductal carcinoma (IDC) and invasive lobular carcinoma (ILC) of the breast on full-field digital mammograms (FFDM).
Methods and Materials: 110 screen detected and histopathologically verified breast cancers (27 DCIS, 73 IDC, 10 ILC) imaged with FFDM in standard views were included in this study. For each lesion, a region of interest (ROI) was manually defined, which covered the lesion as well as a rim (1cm width) of normal-appearing breast tissue around the lesion in the view, where the lesion was depicted in largest diameter. TF derived from the grey-level histogram, co-occurrence matrix (COC), run-length matrix (RLM), absolute gradient (AG), autoregressive model (ARM) and wavelet transform were calculated for the ROIs. Fisher coefficients were calculated to determine which TF were best-suited for distinguishing between DCIS, IDC and ILC. Lesion classification was performed using linear discriminant analysis in conjunction with a k-nearest neighbour classifier, based on the combination of the 10 TF with the highest Fisher coefficients. Classification accuracy was used as the primary outcome measure.
Results: The accuracy of texture-based lesion classification was 84.33% (70 of 83 lesions) for IDC vs. ILC, 81.1% (30 of 37 lesions) for ILC vs. DCIS, but only of 70 % (70 of 100 lesions) for IDC vs. DCIS.
Conclusion: TF derived from FFDM may be of value for differentiating between ILC and IDC, and ILC and DCIS, but of limited value for differentiating between IDC and DCIS.
Methods and Materials: 110 screen detected and histopathologically verified breast cancers (27 DCIS, 73 IDC, 10 ILC) imaged with FFDM in standard views were included in this study. For each lesion, a region of interest (ROI) was manually defined, which covered the lesion as well as a rim (1cm width) of normal-appearing breast tissue around the lesion in the view, where the lesion was depicted in largest diameter. TF derived from the grey-level histogram, co-occurrence matrix (COC), run-length matrix (RLM), absolute gradient (AG), autoregressive model (ARM) and wavelet transform were calculated for the ROIs. Fisher coefficients were calculated to determine which TF were best-suited for distinguishing between DCIS, IDC and ILC. Lesion classification was performed using linear discriminant analysis in conjunction with a k-nearest neighbour classifier, based on the combination of the 10 TF with the highest Fisher coefficients. Classification accuracy was used as the primary outcome measure.
Results: The accuracy of texture-based lesion classification was 84.33% (70 of 83 lesions) for IDC vs. ILC, 81.1% (30 of 37 lesions) for ILC vs. DCIS, but only of 70 % (70 of 100 lesions) for IDC vs. DCIS.
Conclusion: TF derived from FFDM may be of value for differentiating between ILC and IDC, and ILC and DCIS, but of limited value for differentiating between IDC and DCIS.
