Automated breast mass detection in 3D reconstructed tomosynthesis volumes: a featureless approach

Abstract

The purpose of this study was to propose and implement a computer aided detection (CADe) tool for breast tomosynthesis. This task was accomplished in two stages-a highly sensitive mass detector followed by a false positive (FP) reduction stage. Breast tomosynthesis data from 100 human subject cases were used, of which 25 subjects had one or more mass lesions and the rest were normal. For stage 1, filter parameters were optimized via a grid search. The CADe identified suspicious locations were reconstructed to yield 3D CADe volumes of interest. The first stage yielded a maximum sensitivity of 93% with 7.7 FPs/breast volume. Unlike traditional CADe algorithms in which the second stage FP reduction is done via feature extraction and analysis, instead information theory principles were used with mutual information as a similarity metric. Three schemes were proposed, all using leave-one-case-out cross validation sampling. The three schemes, A, B, and C, differed in the composition of their knowledge base of regions of interest (ROIs). Scheme A's knowledge base was comprised of all the mass and FP ROIs generated by the first stage of the algorithm. Scheme B had a knowledge base that contained information from mass ROIs and randomly extracted normal ROIs. Scheme C had information from three sources of information-masses, FPs, and normal ROIs. Also, performance was assessed as a function of the composition of the knowledge base in terms of the number of FP or normal ROIs needed by the system to reach optimal performance. The results indicated that the knowledge base needed no more than 20 times as many FPs and 30 times as many normal ROIs as masses to attain maximal performance. The best overall system performance was 85% sensitivity with 2.4 FPs per breast volume for scheme A, 3.6 FPs per breast volume for scheme B, and 3 FPs per breast volume for scheme C.

Figure 1

Stage 1—the filtration and ROI extraction or the high-sensitivity, low specificity stage of the CADe algorithm.

Figure 2

(a) 25 CADe suspicious locations in 2D for subject 33 (b). Reconstructed CADe suspicious locations using the images in “a.” Significant out of plane blur is observed in Z direction. (c) A 256×256 ROI centered at the X,Y location at the depth with the sharpest focus is extracted from the FBP reconstructed volumes and shown in “b.”

Figure 3

Composition of knowledge base of false positive reduction stage of the CADe algorithm (a) scheme “A” KB composition (b) schemes “A” and “B” composition.

Figure 4

Sensitivity as a function of the two filter parameters for stage 1 of the algorithm. The combination marked by the “+” was chosen, yielding 93% sensitivity with 7.7 FPs∕breast volume.

Figure 5

The figure of merit, ROC AUC, is plotted as a function of increasing number of FP ROIs in the system.

Figure 6

The figure of merit, ROC AUC is plotted as a function of increasing number of normal ROIs in the system.

Figure 7

(a) Nonparametric ROC curves of the central slice classifier for schemes A, B, and C. (b) Partial ROC curves for sensitivity greater than 0.9 for the three schemes.

Figure 8

System FROCs. Prior to FP reduction, the system performance was at 93% sensitivity with 7.7 FPs per breast volume. Final system performances for the three schemes are depicted for the central slice classifiers.

Figure 9

(a) Slice 41 prior to FP reduction, (b) Slice 41 after FP reduction, (c) Slice 21 prior to FP reduction, (d) Slice 21 after FP reduction. Subject 122 had biopsy confirmed carcinoma. While this subject had 6 FPs in total from stage 1 of the CADe algorithm, only reconstructed slices 41 and 21 are shown in this figure for illustration. After setting the threshold for scheme A central slice classifier to operate at 91.5% sensitivity, we are able to eliminate the FP in slice 21. However, the FP in slice 41 survives along with the TP.