Breast mass classification based on opto-acoustic features using data mining

This poster is published under an open license. Please read the disclaimer for further details.

Congress:

ECR 2014

Poster Number:

C-0926

Type:

Scientific Exhibit

Keywords:

Breast, Ultrasound, Biopsy, Cancer

Authors:

A. T. Stavros¹, P. Lavin¹, D. Schoenfeld², M. J. Ulissey¹; ¹San Antonio, TX/US, ²Boston, MA/US

DOI:

10.1594/ecr2014/C-0926

DOI-Link:

https://dx.doi.org/10.1594/ecr2014/C-0926

Fig. 1: Displays an IDC grade 2 with numerous internal red vessels, internal red blush,...

Fig. 2: Displays a Fibroadenoma with normal polar branching feeding and draining...

Table 1: Depicts the five classification methods used to classify between benign and...

Results

The study population consisted of 29 fibroadenomas (FAs) as well as 10 IDC 1, 11 IDC 2, and 18 IDC 3 cases; 66 cases were evaluated for the classification analyses. Results were consistent and favorable for Imagio; there was 100% sensitivity with 72% to 79% specificity (Table 1). KNN and SVM performed the best while LR performed the worst. Findings support the stand-alone biopsy-sparing potential for Imagio. In addition, the probability of malignancy for these methods decreased depending on the variables with negative coefficients and increased with those with positive ones. The coefficients for BA and DV were statistically significant. The importance of these coefficients for the other methods was consistent. The probability of malignancy further depended on the internal tumor features but not on the external features (Figure 1 and 2).

OA Total (two-sided p=3x10-8) and OA Total Internal (two-sided p=6x10-9) were highly significant, but not OA Total External.
OA Total and OA Total Internal Features were highly related to histology with lower scores for FAs increasing progressively to highest scores for IDC 3.
All three internal OA features were highly significant: (1) DV (two-sided p=10-7), BO (two-sided p=3x10-8), and BA (two-sided p=5x10-6).
The cumulative contribution of each internal feature to the total internal score was also assessed for each histopathology subgroup to confirm independent contribution to the OA Total Internal Score; pairwise correlations explained no more than 56% of the total variance.