High-resolution microCT to assess breast microcalcification morphometry by histologic lesion subtype and radiologic classification

Abstract

Purpose:

To assess associations of 3D morphometry of breast calcifications, determined using high-resolution microCT, with underlying histopathology and evaluate the potential for classifying tissue calcifications as an aid in optimizing identification and diagnosis of the most severe pathology in a biopsy.

Methods:

We compared morphometry of calcifications in tissue blocks (n=156) of benign breast disease (n=74), ductal carcinoma in situ (n=39), and ductal carcinoma in situ with invasive breast cancer (n=43) using (10μm) microCT. Matched sections were reviewed microscopically to define lesion(s) in which calcifications were embedded. 3D reconstructions of calcifications were reviewed independently by two masked radiologists to assess the need for biopsy and calcification morphology. Calcification morphometry was compared to pathologic diagnosis using linear mixed models.

Results:

A total of 12,216 calcifications (IQR 9–66 per specimen) were analyzed. Individual breast cancer-associated calcifications were significantly larger than benign breast disease-associated calcifications (p_adjusted=0.012) and calcification volume was positively associated with grade of ductal carcinoma in situ (P_trend = 0.009). Structure model index revealed more cylindrical shape in breast cancer calcifications versus benign breast disease calcifications (p_adjusted=0.008). Re-analysis by the specific microscopic lesion containing the calcification, rather than the most severe diagnosis per biopsy, strengthened associations. Radiologists agreed on biopsy recommendation in 92% of microCT images, while achieving sensitivities of 53.1% and 54.3%. and specificities of 50.0% and 54.1%.

Conclusion:

Our analysis provides proof-of-concept that morphometry of tissue calcifications varies by lesion type, suggesting that future studies may enable development of a pathologic classification linked to diagnosis and mammographic findings.

Background:

Mammographic detection and biopsy of groups of breast microcalcifications (calcifications) results in the diagnosis of up to 90% of ductal carcinoma in-situ (DCIS) lesions and 30–50% of small invasive breast cancers (BCs)[1]. Furthermore, identification of multifocal mammographic calcifications may predict increased risk of developing BC[2]. However, analysis of a cohort of 4,819 women with benign breast disease (BBD) identified calcifications in over 40% of biopsies[3]. Thus, evaluation of lesions associated with mammographic calcifications contributes to early detection of BC while yielding many “false positive” biopsies. Although radiologists apply Breast Imaging Reporting and Data System (BI-RADS) terminology[1, 4, 5] to classify the morphology and distribution of calcifications and to define the need for biopsy, a similar pathologic classification of calcifications is lacking, limiting verification that suspicious mammographic calcifications and related pathology have been viewed microscopically.

Mammographic groups of calcifications may correspond to microscopic areas of BBD, DCIS, and invasive BC, which often co-occur within a single biopsy. Thus, identifying calcifications microscopically may not ensure that the most concerning mammographic lesion has been visualized in routinely prepared 5-μm tissue sections, which reflect only a tiny fraction of the sample. Deeper histologic sectioning of calc-targeted biopsies has been shown to reveal calcifications in 24.6% of cases and more severe pathology in 5.8% of cases in an analysis of 916 consecutive biopsies,[6] and repeat biopsies of radiologic calcs increased BC detection by 14% in a study of 171 women.[7] Thus, we hypothesize that development of a taxonomy of calcifications that is linked to pathologic diagnosis and ultimately to the BI-RADS classification, may improve radiologic-pathologic correlation.

Linking mammographic calcifications with pathologic calcifications is challenging because of differences in scales[1, 8–10]. Screening mammography can detect calcifications larger than 100–200 μm[11], with greater resolution achievable in magnification views, whereas microscopy can detect smaller calcifications with localization to the lesion in which they are embedded. However, routinely prepared 5-μm sections reveal a small percentage of calcifications removed, approximate morphology of calcifications only in 2D, are prone to artifacts during processing, and obscure the natural grouping and distribution on mammography. Sampling concerns are exacerbated among patients with large groups of calcifications or multifocal calcifications that are sampled but not entirely removed by biopsy. High-resolution micro-computed tomography (microCT) has been proposed as an adjunctive method to histopathologic assessment[12] that can generate 3D renderings of calcifications at the 1.2–35μm scale[12–16], with maintenance of structure, thereby complementing histopathologic examination in determining that concerning calcifications are evaluated (Figure 1).

Fig. 1: Breast microcalcifications in mammography, microCT, and histology.

a. Mammographic magnification view of grouped pleomorphic calcifications recommended for biopsy (a1); associated histopathology section with H&E stain highlighting calcifications located in high grade DCIS with comedo-necrosis (a2); microCT image of calcifications located in grade 3 DCIS, capturing the 3D structure of grouped breast calcifications and casting of branching duct (arrow) (a3). B. Spiculated mass with associated amorphous and pleomorphic calcifications (b1) and scattered distribution of calcifications in microCT (b2) and calcifications in stroma associated with invasive BC (b3). C. Suspicious pleomorphic calcifications on mammography(c1), with 3D distribution exemplified on microCT (c2), and corresponding to atypical ductal hyperplasia on histology(c3).

The use of microCT to evaluate breast calcifications was recently reviewed[17] and additionally summarized here (see Supplementary Table 1, Online Resource 1)[12, 15, 18–27]. Studies published from 2011–2024, including from 11–94 tissue samples have identified morphologic features of calcifications that distinguish biopsies diagnosed as BBD versus BC. A prior report found that using machine learning to classify individual calcifications as BBD or BC-associated achieved a sensitivity of 61.2% and specificity of 89.8% (AUC = 0.812) and entire biopsies as BBD or BC resulted in sensitivity 86.5%, specificity 81.4%, and accuracy 84.0%)[18]; however, large samples from diverse sources are required to validate this promising result. Although microCT calcification morphology has been compared to the most severe diagnoses in biopsies, studies have been limited in size and have not related data to specific lesions in which they are embedded, defined microscopically. Further, calcs are histologically identified in multiple lesions in a single biopsy. To extend prior research, we assess microCT features of calcifications in a range of pathologic lesions, with histopathologic review to specify calcification-associated lesions. Our objective is to demonstrate the link between 3D morphologic features of histologic breast calcifications and the specific microscopic lesion in which they are embedded. As a step towards radiologic correlation, we evaluated classification of microCT images by breast radiologists.

Methods

Participants

Surgical biopsies from the Mayo Clinic BBD cohort including biopsies of BBD and DCIS ± invasive BC were studied (n=156), including excisional biopsies (n=77; 49.4%), wide local excisions (n=48; 30.8%), and mastectomies (n=31, 19.9%) performed at Mayo Clinic, Rochester (1992–2021)[28–30]. Biopsies with calcifications identified on 5-μm H&E-stained sections and/or on microCT evaluation were included. A stratified sample with final pathologic diagnoses of BBD (n=74; subclassified as non-proliferative (NP) n=24; proliferative disease without atypia (PDWA) n=25; or atypical hyperplasia (AH); n=25), DCIS (n=39; Grade 1 (n=15); Grade 2 (n=12); and Grade 3 (n=12)), and invasive BC with DCIS (n=43; Grade 1 (n=15); Grade 2 (n=15); and Grade 3 (n=13)) were included. Biopsies were selected without examination of calcification morphology. Association(s) of calcifications by lesion type(s) was defined by microscopic review of corresponding 5um H&E-stained sections of the matched specimens by an experienced pathologist (MES). A subset of patients (n=16) contributed both a BBD biopsy and BC specimen occurring years later, which we considered independently. Use of human breast biopsies was approved by Mayo Clinic IRB.

MicroCT analysis

MicroCT images of the formalin-fixed paraffin-embedded diagnostic tissue blocks were collected on a Bruker SkyScan 1276 microCT scanner (Bruker, Belgium), with operating voltage of 70kV, 0.5 mm Al filter and 0.2-degree rotation step. Data were reconstructed with a 10 μm isotropic cubic voxel size. Images were segmented to define calcifications and image analysis was performed using SkyScan analysis CTAn V1.20.8.0 (Bruker, Belgium) software. Objects less than 3 voxels in volume or with a structural thickness measuring 1 or 2 pixels were considered artifact and removed. Object parameters were calculated for each individual calcification: volume (mm³), surface area (mm²), structural thickness (average in mm), major diameter (mm), structural model index (SMI), and sphericity. SMI is a continuous measure of shape ranging from 0 (perfect plate) to 3 (perfect cylinder) to 4 (perfect sphere).

Radiologist Assessment

3D reconstructed images of calcification objects were rendered utilizing SkyScan CTVox Software Version 3.3.1 (Bruker, Belgium). Two breast radiologists, masked to clinical data including radiology reports, independently reviewed the 3D images to identify and apply BI-RADS classification to the most suspicious calcification group and assessed the need for biopsy. Of note, microCT reconstructions only show calcifications in detail. Joint review, masked to diagnosis, was completed to reach consensus on recommendation for biopsy in cases of disagreement (n=12).

Statistical methods

Associations of demographic characteristics with tissue type were examined using t-tests for continuous variables and chi-square tests or, if any expected cell counts were less than 5, Fisher exact tests for categorical variables. Analysis of calcification morphometry was performed in two ways, most severe diagnosis per block and in relation to the specific lesion(s) containing calcifications defined microscopically in 2D histopathologic analysis of H&E-stained sections. Primary analyses compared continuously distributed features of calcifications in BBD and BC, defined as DCIS with or without associated invasive BC. The median of calcification volume and SMI across calcifications within a specimen were calculated and the resulting specimen medians and inter-quartile ranges (IQRs) were estimated from these values. Calcification volume and SMI and all other calcification morphometric features were then transformed using an inverse normal (probit) transformation and linear mixed models were developed, fitting the transformed morphologic measures of interest as the outcome variable, tissue type as the exposure variable, and specimen as a random intercept term using a compound symmetry variance-covariance matrix. Resulting parameter estimates can be interpreted as the change in normally distributed z-scores for a given category relative to the referent group. Multivariable models were adjusted for potential confounders of calcification morphology and pathology diagnosis, including age of patient at surgery, number of calcifications in the specimen, and screening modality (plain film versus full-field digital mammography). For simplicity, linear mixed models of volume and SMI are shown while models of other imaging parameters are included in the supplementary tables (see results). Multiple sensitivity analyses were performed: (1) restricting to calcifications with a major diameter of ≥70 μm, approximating the threshold for mammographic visualization, (2) restricting to biopsies performed for calcifications, (3) excluding biopsies in which calcifications were found by microCT only, and (4) removing BBD biopsies provided by patients who also contributed a BC tissue. Heterogeneity among calcification features (quantified by standard deviation of morphologic features across calcifications within a specimen) were compared across demographic and clinical characteristics using a Kruskal-Wallis test.

Agreement between two masked independent radiologists on need for biopsy, defined as biopsy recommended or not recommended, and most suspicious BI-RADS morphology term based on review of microCT images was calculated, ordered by decreasing level of suspicion (fine linear/fine linear branching, fine pleomorphic, amorphous, coarse heterogenous; all benign terms were grouped: coarse/“popcorn-like”, large rod-like, round, rim, dystrophic, milk of calcium, suture). Sensitivity and specificity of biopsy recommendations were calculated based on histopathologic diagnosis of the block, defined as BBD versus BC, and repeated by calcification-containing lesion type, when uniquely specified. Linear mixed models were utilized to compare morphologic characteristics of calcifications in tissues recommended for biopsy compared to those not recommended for biopsy and between correctly classified and misclassified biopsies. BlueSky Statistics software v. 10.3.1 (BlueSky Statistics LLC, Chicago, IL, USA) and RStudio ^® v. 2024.09.0 (Posit Software PBC) were used to conduct analyses.

Results

Cohort clinical characteristics and demographics

Overall, 156 biopsies from 140 patients, 96.2% non-Hispanic White Women, with median age 60.3 (IQR 51.6, 69.0) years, were studied. BC patients were older than those with BBD (Table 1). Suspicious mammographic calcifications prompted the excision in 66% of both BBD and BC patients, although all contained calcifications either microscopically and/or on microCT imaging.

Table 1:

Demographic features and indication for excision

	All (n=156)	BC (n=82)	BBD (n=74)	P-value
Age	60.3 (51.6, 69.0)	62.3 (55.0, 70.8)	57.6 (46.7, 67.6)	0.0131
Race				0.1202
Asian	1 (0.6%)	0 (0.0%)	1 (1.4%)
White	151 (96.2%)	82 (98.8%)	69 (93.2%)
Unknown	5 (3.2%)	1 (1.2%)	4 (5.4%)
Suspicious or indeterminate calcifications on mammography				0.7733
Yes	97 (66.4%)	51 (65.4%)	46 (67.6%)
No	49 (33.6%)	27 (34.6%)	22 (32.4%)
NA	10	4	6

The demographic information and indication for biopsy are compared between tissues containing benign (BBD=benign breast disease) and malignant breast disease (BC=breast cancer). Median (IQR) or N (%) provided.

¹T-test

²Fischer’s exact test

³Pearson’s chi-squared test

Calcification morphometry versus pathologic diagnosis per block

A total of 12,216 calcifications (IQR 9–66 per specimen) were identified by microCT. In unadjusted analyses, median calcification volume in biopsies diagnosed as BC (median of median across calcifications per specimen) was 2.76 ×10⁻⁴ mm³ compared with 2.21×10⁻⁴ mm³ for BBD (p=0.012; p-values are from linear mixed models) and greater in high-grade (4.40 ×10⁻⁴ mm³) than in low-grade BC (2.72 ×10⁻⁴ mm³, p=0.049), with increased statistical significance after adjustment for potential confounders (Table 2A, for full model see Supplementary Table 2, Online Resource 1). Grade 3 DCIS included largest calcification volumes (6.38 × 10⁻⁴ mm³, p_adjusted=0.028). After adjustment for potential confounders, among DCIS, increasing DCIS grade was associated with increased individual calcification volume (p_trend = 0.009). Multivariable analyses of SMI demonstrated that calcifications in BC are more cylindrical than in BBD (p=0.008, Table 2B, for full model see Supplementary Table 3, Online Resource 1). In analyses restricted to BBD, individual calcification volume in PDWA was smaller (1.86 ×10⁻⁴ mm³) than in NP (2.89 ×10⁻⁴ mm³, p_adjusted=0.030). Other comparisons for calcification volume and SMI among both BBD and BC specimens were not significantly different. Multivariable models assessing other morphometric features identified similar findings to volume for size (see Supplementary Tables 4–6, Online Resource 1) and to SMI for shape (see Supplementary Table 7, Online Resource 1). Sensitivity analyses, restricted to calcifications with a major diameter of ≥ 70 μm (see Supplementary Table 8, Online Resource 1), calcification-targeted biopsies (see Supplementary Table 9, Online Resource 1), biopsies with calcifications on histologic sectioning (see Supplementary Table 10, Online Resource 1), and unique patients (see Supplementary Table 11, Online Resource 1) identified similar results compared to overall analysis.

Table 2:

Linear mixed models of calcification morphometry by most severe pathologic lesion.

A. Volume		Univariable model			Multivariable model1
Variable	Median (IQR) (mm3)2	Parameter Estimate3	Std. Error	p-value	Parameter Estimate3	Std. Error	p-value
BBD (n=74)	2.21×10−4 (1.90 ×10−4)	0	(REF)		0	(REF)
BC (n=82)	2.76×10−4 (3.30 ×10−4)	0.199	0.077	0.010	0.247	0.096	0.012
NP (n=24)	2.89 ×10−4 (2.75 ×10−4)	0	(REF)		0	(REF)
PDWA (n=25)	1.86 ×10−4 (1.36 ×10−4)	−0.279	0.134	0.041	−0.294	0.133	0.030
AH (n=25)	2.19 ×10−4 (2.39 ×10−4)	−0.18	0.135	0.187	−0.068	0.144	0.637
Low grade BC (n=57)	2.72 ×10−4 (2.33 ×10−4)	0	(REF)		0	(REF)
High grade BC (n=25)	4.40 ×10−4 (6.60 ×10−4)	0.224	0.112	0.049	0.246	0.118	0.041
DCIS only (n=39)	3.40 ×10−4 (5.42 ×10−4)	0	(REF)		0	(REF)
DCIS with invasive BC (n=43)	2.62 ×10−4 (2.62 ×10−4)	−0.089	0.105	0.396	−0.095	0.112	0.403
DCIS Grade 1 (n=15)	3.34 ×10−4 (2.30 ×10−4)				0	(REF)
DCIS Grade 2 (n=12)	2.73 ×10−4 (5.31 ×10−4)	0.015	0.209	0.943	0.148	0.218	0.501
DCIS Grade 3 (n=12)	6.38 ×10−4 (1.22 ×10−4)	0.29	0.21	0.176	0.537	0.233	0.028
Invasive BC Grade 1 (n=15)	2.62 ×10−4 (2.41 ×10−4)	0	(REF)		0	(REF)
Invasive BC Grade 2 (n=15)	2.62 ×10−4 (2.10 ×10−4)	0.021	0.143	0.886	0.045	0.152	0.767
Invasive BC Grade 3 (n=13)	2.48 ×10−4 (3.19 ×10−4)	0.17	0.156	0.285	0.177	0.163	0.287
B. SMI		Univariable model			Multivariable model1
Variable	Median (IQR)2	Parameter Estimate4	Std. Error	p-value	Parameter Estimate 4	Std. Error	p-value
BBD (n=74)	3.22 (0.15)	0	(REF)		0	(REF)
BC (n=82)	3.20 (0.15)	−0.048	0.085	0.575	−0.265	0.104	0.012*
NP (n=24)	3.25 (0.16)	0	(REF)		0	(REF)
PDWA (n=25)	3.18 (0.16)	−0.115	0.155	0.459	−0.111	0.158	0.484
AH (n=25)	3.25 (0.13)	0.095	0.156	0.543	0.01	0.172	0.955
Low grade BC (n=57)	3.20 (0.21)	0	(REF)		0	(REF)
High grade BC (n=25)	3.20 (0.11)	−0.094	0.126	0.455	−0.056	0.125	0.653
DCIS only (n=39)	3.22 (0.14)	0	(REF)		0	(REF)
DCIS with invasive BC (n=43)	3.19 (0.19)	−0.119	0.115	0.303	−0.009	0.116	0.939
DCIS Grade 1 (n=15)	3.32 (0.15)				0	(REF)
DCIS Grade 2 (n=12)	3.21 (0.12)	−0.163	0.208	0.439	−0.29	0.219	0.194
DCIS Grade 3 (n=12)	3.20 (0.11)	−0.154	0.209	0.468	−0.35	0.234	0.144
Invasive BC Grade 1 (n=15)	3.09 (0.17)	0	(REF)		0	(REF)
Invasive BC Grade 2 (n=15)	3.21 (0.19)	0.282	0.185	0.137	0.139	0.174	0.431
Invasive BC Grade 3 (n=13)	3.19 (0.06)	0.021	0.198	0.916	0.025	0.183	0.89

Linear mixed models (A) object volume and (B) structure model index by most severe lesion present in the specimen without consideration of calcification location are shown.

Linear mixed models fit probit-transformed morphometry variable of interest as the outcome and histology characteristic of interest as exposure, and accounting for multiple calcification measurements per patient by fitting a per-patient random intercept term with compound symmetry variance-covariance matrix. Complete univariable and multivariable models are available (see Supplementary Figure 2A and 2B, Online Resource 1). DCIS = ductal carcinoma in situ, invasive BC = invasive breast cancer, BC = DCIS with and without invasive BC, BBD = benign breast disease, NP = nonproliferative, PDWA = proliferative disease without atypia, AH = atypical hyperplasia.

¹Additional adjustment for age, number of calcifications per sample, and screening modality

²Median calcification volumes and SMIs were calculated for each specimen and the median and IQR of these values are shown here.

³Estimated change in probit-transformed object volume for category of interest compared to the referent group

⁴Estimated change in probit-transformed object structure model index depending on the table for category of interest compared to the referent group

Calcification morphometry by pathologic lesion

The location of calcifications within specific histopathologic lesions was available for 150 specimens, enabling discrimination of scans with calcifications in BBD from those with calcifications in BBD and DCIS with or without invasive BC. For biopsies considered DCIS in overall histologic assessment, calcifications were located in DCIS only in 56.4% (n=22), BBD only in 7.7% (n=3), and both DCIS and BBD in 35.9% (n=14). In biopsies with invasive BC, calcifications were located in invasive BC only in 7.9% (n=3), in DCIS only in 31.6% (n=12), BBD only in 10.5% (n=4), both BBD and BC in 31.6% (n=12), and in both DCIS and invasive BC in 18.4% (n=7). Calcification volumes in BC tissues (2.89 ×10⁻⁴ mm³) were larger than those located in BBD (2.21 ×10⁻⁴ mm³; p_adjusted=0.002, Table 3), and intermediate (although smallest in median analysis) when both BBD and BC were present in the block in mixed-model analysis. Although based on limited samples sizes, calcification volumes were highest when calcifications were restricted to DCIS (4.18 ×10⁻⁴ mm³), smaller in DCIS with invasive BC (1.87 ×10⁻⁴ mm³), and smallest when present only in invasive BC (0.97 ×10⁻⁴ mm³, p_adjusted=0.007). When analyzed by most severe lesion containing calcifications, volumes in DCIS grade 3 (4.98 ×10⁻⁴ mm³) were greater than in DCIS grade 1 (3.09 ×10⁻⁴ mm³) or grade 2 (2.48 ×10⁻⁴ mm³; p_trend=0.030). SMI differed for blocks containing BC (3.20) versus those diagnosed as BBD (3.22, p_adjusted=0.005) and for blocks with DCIS plus invasive BC (3.32) versus DCIS without invasion (3.20, p_adjusted=0.014). Calcification morphometry was similar when present in BBD or in “normal lobules”.

Table 3:

Linear mixed models of calcification morphometry by surrounding lesion.

A. Volume		Univariable model			Multivariable1
Variable	Median (IQR) (mm3)2	Parameter Estimate3	Std. Error	p-value	Parameter Estimate3	Std. Error	p-value
BBD only (n=80)	2.21 ×10−4 (2.03 ×10−4)	0	(REF)		0	(REF)
BC only (n=44)	2.89 ×10−4 (5.13 ×10−4)	0.267	0.092	0.004	0.357	0.116	0.002
Mixed BBD and BC (n=26)	2.02 ×10−4 (3.40 ×10−4)	0.112	0.106	0.29	0.189	0.123	0.128
NP (n=38)	2.63 ×10−4 (2.90 ×10−4)	0	(REF)		0	(REF)
PDWA (n=17)	2.56 ×10−4 (1.29 ×10−4)	−0.078	0.121	0.521	−0.061	0.118	0.609
AH (n=14)	2.06 ×10−4 (1.58 ×10−4)	−0.194	0.125	0.127	−0.009	0.118	0.951
Normal tissue (n=9)	1.06 ×10−4 (8.17 ×10−4)	−0.186	0.156	0.238	−0.15	0.142	0.350
Low grade BC (n=49)	2.48 ×10−4 (2.59 ×10−4)	0	(REF)		0	(REF)
High grade BC (n=21)	5.08 ×10−4 (9.49 ×10−4)	0.386	0.12	0.002	0.438	0.127	0.001
DCIS only (n=34)	4.18 ×10−4 (8.23 ×10−4)	0	(REF)		0	(REF)
DCIS and invasive BC (n=7)	1.87 ×10−4 (1.14 ×10−4)	−0.506	0.191	0.01	−0.552	0.197	0.007
Invasive BC only (n=3)	0.97 ×10−4 (0.97 ×10−4)	−0.798	0.305	0.011	−0.891	0.317	0.007
Malignant and benign (n=26)	2.02 ×10−4 (3.40 ×10−4)	−0.282	0.118	0.019	−0.296	0.12	0.017
DCIS Grade 1 (n=18)	3.09 ×10−4 (2.17 ×10−4)	0	(REF)		0	(REF)
DCIS Grade 2 (n=17)	2.48 ×10−4 (5.18 ×10−4)	−0.043	0.162	0.792	0.001	0.164	0.997
DCIS Grade 3 (n=20)	4.98 ×10−4 (1.01 ×10−4)	0.282	0.157	0.077	0.372	0.166	0.030
Invasive BC Grade 1 (n=6)	1.34 ×10−4 (2.86 ×10−4)	0	(REF)		0	(REF)
Invasive BC Grade 2 (n=8)	1.96 ×10−4 (8.17 ×10−4)	0.053	0.169	0.759	0.211	0.214	0.35
Invasive BC Grade 3 (n=1)	5.64 ×10−4 (NA)	0.632	0.389	0.13	0.766	0.42	0.101
B. SMI					Multivariable model1
Variable	Median (IQR)2	Parameter Estimate4	Std. Error	p-value	Parameter Estimate4	Std. Error	p-value
BBD (n=80)	3.22 (0.165)	0	(REF)		0	(REF)
BC (n=44)	3.20 (0.134)	−0.059	0.102	0.562	−0.351	0.123	0.005
Mixed BBD and BC (n=26)	3.20 (0.175)	0.025	0.119	0.836	−0.229	0.133	0.086
NP (n=38)	3.20 (0.218)	0	(REF)		0	(REF)
PDWA (n=17)	3.23 (0.141)	0.038	0.163	0.817	0.032	0.166	0.846
AH (n=14)	3.25 (0.120)	0.363	0.172	0.038*	0.229	0.203	0.264
Normal tissue (n=9)	3.19 (0.214)	0.166	0.213	0.44	0.106	0.229	0.646
Low grade BC (n=49)	3.21 (0.184)	0	(REF)		0	(REF)
High grade BC (n=21)	3.20 (0.118)	−0.109	0.134	0.415	−0.177	0.133	0.187
DCIS only (n=34)	3.20 (0.117)	0	(REF)		0	(REF)
DCIS and invasive BC (n=7)	3.32 (0.109)	0.581	0.202	0.006	0.506	0.2	0.014
Invasive BC only (n=3)	3.20 (0.066)	−0.137	0.313	0.663	−0.032	0.313	0.919
Malignant and benign (n=26)	3.20 (0.109)	0.17	0.126	0.182	0.202	0.123	0.104
DCIS Grade 1 (n=18)	3.22 (0.138)				0	(REF)
DCIS Grade 2 (n=17)	3.19 (0.129)	−0.087	0.172	0.616	−0.086	0.17	0.614
DCIS Grade 3 (n=20)	3.20 (0.120)	−0.093	0.166	0.579	−0.187	0.172	0.28
Invasive BC Grade 1 (n=6)	3.20 (0.185)	0	(REF)		0	(REF)
Invasive BC Grade 2 (n=8)	3.30 (0.156)	0.194	0.29	0.517	−0.142	0.306	0.654
Invasive BC Grade 3 (n=1)	3.20 (NA)	−0.19	0.597	0.756	−0.472	0.557	0.419

Linear mixed models of (A) object volume and (B) structure model index of calcifications by type of lesion in which they were embedded are shown. Linear mixed models fit probit-transformed morphometry variable of interest as the outcome and histology and type of lesion as exposure, and accounting for multiple calcification measurements per patient by fitting a per-patient random intercept term with compound symmetry variance-covariance matrix. Complete univariable and multivariable models are available (see Supplementary Figure 6A and 6B, Online Resource 1). DCIS only indicates calcification located only in DCIS, pure invasive BC indicates calcifications located only in invasive BC, mixed BC indicates calcifications located in both DCIS and invasive BC. For DCIS and invasive BC grade analysis and BBD histologic subtype analysis, the category represents the most severe lesion in the specimen containing calcifications. DCIS = ductal carcinoma in situ, invasive BC = invasive breast cancer, BC = DCIS with and without invasive BC, BBD = benign breast disease, NP = nonproliferative, PDWA = proliferative disease without atypia, AH = atypical hyperplasia

¹Additional adjustment for age, number of calcifications per sample, and screening modality

²Median calcification volumes and SMIs were calculated for each specimen and the median and IQR of these values are shown here.

³Estimated change in probit-transformed object volume for category of interest compared to the referent group

⁴Estimated change in probit-transformed object structure model index depending on the table for category of interest compared to the referent group

Heterogeneity of calcification morphometry by associated lesion type

Calcifications contained in BC tissues were significantly more heterogeneous in volume (measured by standard deviation of volume; p=0.037, see Supplementary Table 12, Online Resource 1). In biopsies containing DCIS and invasive BC, heterogeneity in shape was associated with higher grade (p_trend = 0.037). No other significant differences in volume heterogeneity or shape heterogeneity were identified between groups.

Radiologist review of 3D reconstruction of calcification objects

Calcifications were too small to assess visually in one specimen. Radiologist 1 recommended biopsy for 80 (51.6%) of 155 biopsies, with a sensitivity of 53.1% and a specificity of 50.0%. Radiologist 2 recommended biopsy for 84 (54.2%) biopsies with a sensitivity of 54.3% and specificity of 54.1%. For both reviewers, classification by calcification-associated lesions yielded similar results. Percent agreement on recommendation of biopsy between radiologists was 92.3%. Agreement was 71.4% on the most severe BI-RADS classifier. Calcifications recommended for biopsy were smaller than those not recommended for biopsy (see Supplementary Table 13, Online Resource 1, p=0.014). Calcifications in BCs not recommended for biopsy based on microCT review were significantly larger than calcifications in BBD not recommended for biopsy (p=0.013). Calcifications in BBD recommended for biopsy were smallest. BC-associated calcifications without a recommendation for biopsy were more cylindrical than correctly identified BBD-associated calcifications, after adjustment (0.038). Application of BI-RADS terms to histologic groups is summarized (see Supplementary Table 14, Online Resource 1).

Discussion

Radiological-pathological correlation is crucial for optimal management of suspicious mammographic calcifications, a key sentinel of DCIS and early invasive BC, and a source of many BBD biopsies. Improving the management of calcifications takes on added significance in trials assessing management of atypical ductal hyperplasia and low-grade DCIS with observation, wherein excluding a more severe lesion may be critical [31, 32]. Furthermore, uncertainties about the pathology represented by residual calcifications remaining after excision of atypical ductal hyperplasia, DCIS, or invasive BC may influence decisions about the need for a second surgery. Thus, developing a classification of calcifications in tissues linked to underlying pathology and to radiologic features could improve management.

Challenges in correlating the appearance of calcifications in mammography and pathology include differences in scale, sampling, appearance, and artifacts. Nearly one-third of specimens included in this study contained calcifications in both BBD and DCIS and/or invasive BC, complicating correlations at the tissue level. However, our sub-analysis focusing on tissues containing calcifications restricted to one type of pathologic lesion, highlights that morphometry of calcifications is strongly linked to the pathologic lesion in which they are embedded. This may have greatest importance when calcifications are found in BBD microscopically, but the radiologic target is highly suspicious.

Our findings demonstrate: 1) individual calcification volumes associated with BC are significantly larger than those in BBD; 2) calcification volumes are greater in high-grade versus low-grade BC; and 3) calcifications in BC are more cylindrical than in BBD. As calcifications in DCIS, with or without invasion, drive these associations. Thus, our results may stimulate further research on calcification morphometry and development of guidelines for calcification reporting in pathology clinical practice, considering the frequent co-occurrence of DCIS/IBC and BBD on biopsy[33]. Although radiologists achieved high inter-rater concordance in classifying calcification morphology and the need for biopsy based on review of microCT images, sensitivity, and specificity were low, likely attributable to varying morphologic appearance and distribution on microCT images compared to standard breast magnification mammography as well as a lack of chronicity data (i.e. newly developed calcifications since last exam).

Larger mammographic calcifications are usually considered benign, consistent with the appearance of calcifications in NP BBD, and may be judged as such mammographically and not biopsied[11]. Furthermore, previous research utilizing microCT has identified that calcifications contained in benign vacuum-assisted core biopsies are generally larger [15, 22] than calcifications contained in DCIS and invasive BC biopsies. Unlike previous studies[15, 18, 22], we utilized excisional biopsies to structure an even sample of BBD histologic subtypes increasing the representation of AH and potentially PDWA, which we found had smaller calcifications than NP, the most frequent source of calcifications[3]. Our finding of larger calcifications in BC than BBD was confirmed in sensitivity analyses of biopsies prompted by calcifications. Indeed, we have shown that calcifications in biopsies with BBD alone, with no histologic evidence of DCIS or invasive BC, are significantly smaller than those associated with BC, ensuring that the presence of BC has not influences calcification morphology. In fact, our analysis showed that the co-occurrence of DCIS with BBD was associated with a reduction in calcification volumes per block compared to specimens containing calcifications restricted to DCIS. Thus, we proposed that conclusions about relative size of calcifications in BBD versus DCIS/IBC will be influenced by the relative number of different BBD subtypes studied, especially NP BBD. Although calcifications in BBD and DCIS overall are significantly different, the heterogeneity within these diagnostic classes results in overlap between characteristics in individual patients limiting discrimination.

Calcifications located in DCIS, the predominant source of malignant calcifications, were significantly larger than calcifications presented in invasive BC only, suggesting that minimally or non-calcifying DCIS may delay detection until a mass containing invasive BC forms. Grade 3 DCIS was associated with greater volume than lower grades and calcifications within BC tissues were significantly more heterogeneous in volume than in benign tissues, consistent with heterogeneity as a mammographic indicator of suspicion[11]. Confirmation of these findings may help exclude potential candidates for conservative management of atypical ductal hyperplasia or low-grade DCIS.

Generally, benign calcifications are identified on mammography by their more regular and spherical shape[11]. In agreement with this established association, we found that calcifications located in BBD were significantly more spherical than BC-associated calcifications. Previous research has not identified significant differences in shape between benign and malignant calcifications[22]. However, our large sample size and adjustment for potential confounders likely improved our ability to discern these differences. Furthermore, we found that the shape of calcifications varied with mammography technique, which is notable because full-field digital mammography is associated with increased detection of DCIS[34], increased detection of BC presenting as grouped calcifications[35] and fewer interval BCs (false-negative screens) presenting as calcifications[36], suggesting improved detection of clinically relevant calcifications. Indeed, the association of increased DCIS grade with elongated, cylindrical shape, consistent with radiologic assessment[11, 37, 38], was strengthened by adjustment for screening era, although not statistically significantly. Adjustment for age and number of calcifications per block did not significantly alter results.

Radiologist assessment of microCT images generally followed mammographic criteria, recommending smaller calcifications for biopsy. Radiologic review was likely limited by the fact that these reviews differ from clinical practice because radiologists were masked to macroscopic distribution within the breast, changes over time, and clinical context. Additionally, image processing removes soft tissue and texture imaging features important to mammographic characterization of breast lesions. However, limited radiologic characterization of microCT imaged calcs may point to difficulties in characterizing mammographic calcifications with uncharacteristic features of BC. Furthermore, these lesions were judged as requiring biopsy clinically and likely skewed towards ambiguous morphology. Nonetheless, interobserver agreement was high, indicating features recognized as concerning were identified similarly between radiologists at a single institution. Future studies assessing whether training with microCT images, and potentially mammographic images, to classify calcifications embedded in histo-pathologically confirmed lesions may improve classification and have value for developing machine learning algorithms for the identification of suspicious calc-associated lesions.

MicroCT is a high-resolution imaging modality capable of evaluating smaller calcifications than are apparent on mammography[15]. Evolution of radiologic and pathologic methods, including novel devices that enable assessment of calcifications in vivo or in tissue specimens may lead to convergence of disciplines in areas requiring radiologic-pathologic correlation, such as management of calcifications. Our study adds a unique analysis of 3D calcification morphology utilizing high-resolution microCT with detailed histopathologic analysis by BBD subtype, microscopic localization of calcifications, radiologists review, and adjustment for potential confounders. Limitations of our study include analysis of almost exclusively white patients residing in the Midwest and treated at a referral center, morphometric analysis limited to a few features not considering spatial relationships of calcifications, and association of calcs with pathologic lesion by histologic evaluation of only clinically collected H&E-stained sections. We have begun to address the limitations inherent in multiple pathologies present per biopsy by performing overall analysis by most severe histologic lesion and by biopsies in which calcifications were located in only one lesion, which generally strengthened the identified associations. Future studies performed with machine learning in conjunction with a validation cohort study design, as well as registration of H&E-stained sections with microCT images and associated mammographic images, may enable development of a pathologic classification of calcifications with value for clinical translation.

Data availability:

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.