Figure 1.
Schematic of our conceptual approach.
Cartoon representing A) a “negative” tumor margin surface corresponding to a mix of normal tissues, and B) a “positive” tumor margin surface with areas of residual tumor (red) at the surface. The cumulative distribution functions in C) and D) are actual optical data from representative margins (sensitive to the relative amount of fat and fibroglandular tissue). In a negative margin (C), there is a distribution of values, which corresponds to the mixture of tissue types in benign (aka negative) margins. When malignancy is present in varying amounts (D), a shift in the optical contrast distribution is observed, due to the disruption in the tissue landscape caused by the increase in cancerous tissue and displacement of normal tissue.
Table 1.
Patient and tumor demographics.
Figure 2.
Relationship between optical parameters and benign breast composition.
A) 50x bicubic interpolated images of [β-carotene], <µs′>, and [β-carotene]/<µs′> from a negative margin. Sites with corresponding histopathology are highlighted with diagnoses of adipose (A) or fibroglandular plus adipose (FG+A). B) Empirical cumulative distribution functions (eCDFs) of the site-level data for fibroglandular (FG), fibroadipose (FA), and adipose (A) sites.
Figure 3.
Relationship between optical parameters and benign breast composition with differences in breast density.
A) Parameter maps from a low density margin; B) Parameter maps from a high density margin; blue indicates higher values of the corresponding variable. C) eCDFs of all measured sites from negative, neoadjuvant-naïve margins, separated by mammographic breast density. P-values were calculated with modified Kolmogorov-Smirnov statistics.
Figure 4.
Analysis of adipose tissue between low and high density breast tissue.
A) Representative H&E micrographs (100x) from all 4 mammographic breast densities (MBD). Cell area and cell density were calculated from an automated image analysis algorithm applied to H&E slides. [β-carotene] and <µs′> were measured via quantitative spectral imaging. B) The adipose sites are from the negative margins of neoadjuvant-naïve patients with a BMI restricted to 25–30. P-values were calculated with a Wilcoxon rank-sum.
Figure 5.
Optical parameters reflect presence of residual disease.
50x bicubic interpolated images of [β-carotene], <µs′>, and [β-carotene]/<µs′> from a A) negative and B) positive margin in 2 different patients with MBD-3. Sites with corresponding histopathology are highlighted with diagnoses of adipose (A), adipose plus fibroglandular (A+FG), and ductal carcinoma in situ (DCIS). C) eCDFs of the pixels from the representative images in panels A and B.
Figure 6.
Optical contrast between negative and positive margins, stratified by mammographic breast density.
A–B) eCDFs of all measured sites from negative and positive, neoadjuvant-naïve margins, separated by radiographic breast density. The P-values indicated correspond to modified Kolmogorov-Smirnov tests between: 1) negative versus positive margins in low density patients and 2) negative versus positive margins in high density patients. C) Difference eCDFs calculated between the positive and negative margin eCDFs of all measured sites for both low and high density breasts. Negative values indicate the negative margin distribution has higher values of the optical variable.
Table 2.
Summary of predictor variables selected by the conditional inference tree model, stratified by mammographic breast density (MBD).
Table 3.
Sensitivity (Se), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV), and classification accuracy (A) of the device and the surgeon.
Table 4.
Number of false negative (FN) and false positive (FP) margins (stratified by margin histology and surgical margin status) and patients (stratified by MBD) calculated from the surgeon performance, as well as, the performance of the device.