Background
Breast conserving surgery (BCS) in combination with radiotherapy has become standard-of-care for treatment of early stage breast cancer, offering equivalent survival to full mastectomy, but with a far less extensive surgical procedure [
1]. However, this equivalency holds true only if resected specimens have negative margins [
2,
3]. In non-palpable cancers, BCS guidance involves pre-operative wire placement [
4,
5] or radioactive seed tumor localization [
6], often followed by specimen mammography [
7,
8]. Yet, 20–40% of patients who undergo BCS require re-excision because of a pathologically positive or close margin [
9]. Additionally, substantial variability exists among surgeons and institutions, with a reported re-excision rates in the range of 0–70% and 1.7–20.9%, respectively [
9,
10], highlighting the clinical need for better surgical guidance during BCS procedures.
The pre-clinical imaging technology, micro-Computed Tomography (micro-CT), offers three dimensional (3D) image reconstruction with sub-millimeter resolution but at much smaller fields of view than clinical CT. While this technology has been utilized extensively in pre-clinical specimen and animal studies [
11‐
13], it has not been widely used to image surgically resected breast tissues [
14]. Tang et al. have motivated the potential value of this technology in assessing shaved cavity margins [
15] and they have also shown it to measure the largest tumor dimension in resected BCS specimens accurately [
16]. However, the diagnostic value of micro-CT in assessing whole resected BCS specimens has not been investigated. Accordingly, the feasibility and potential diagnostic value of micro-CT was evaluated as a surgical guidance tool for BCS by performing an initial correlation of micro-CT analysis to the final histopathologic diagnosis, and qualitatively comparing specimen micro-CT to single-projection specimen mammography.
Materials and methods
Overview of specimen imaging protocol
From May of 2017 to January of 2018, a cohort of 32 BCS specimens was imaged prospectively under a HIPPA compliant, Institutional Review Board (IRB) approved observational protocol. Patients undergoing consented and elective BCS were considered for this study. After each specimen was surgically removed and inked to record its orientation [
17], it was placed on a radiolucent grid (AccuGrid, Beekley Medical, Bristol, CT), sealed in a plastic bag, and underwent single-projection specimen mammography to confirm the primary lesion and surgical clip were removed from the patient. The sealed specimen was then sent to the Pathology specimen grossing laboratory. Upon arrival, each specimen was removed from the plastic bag, then scanned with the micro-CT in the same orientation as single-projection specimen mammography, and subsequently returned to the pathologists’ assistant for standard gross analysis and histological processing. Only single-projection specimen mammography was available for this study. Micro-CT scans were analyzed retrospectively by an experienced breast radiologist (R.A.Z) blinded to the pathological diagnosis. Participant consent was waived by the IRB because the study did not interfere with standard-of-care nor did it disclose or analyze protected health information (PHI).
Micro-CT imaging device and reconstructions
Upon arrival in Pathology, specimens were compressed between two acrylic plates and imaged with a micro-CT scanner (IVIS SpectrumCT, PN 128,201, PerkinElmer, Hopkington, MA), which was physically located in the specimen grossing laboratory in order to minimize tissue handling time. Scanning was performed at X-ray settings of 50 kVp and 1 mA with a 440 µm aluminum filter. The X-ray source-detector pair was fixed, with a 50 µm focal spot size. The specimen was rotated through 360° and received a 100 ms exposure at each 0.5° increment (72 s acquisition). Standard filtered back projection (FBP) was used to reconstruct an image volume of 12 × 12 × 3 cm
3 with 0.15 mm cubic voxels (~ 2 min. reconstruction time). The X-ray detector (3072 × 864 pixels) was binned down by a factor of 4 (1024 × 216 pixels) in order to increase the signal to noise ratio (SNR) and decrease the total scan time. A previous publication reported the micro-CT to have a resolution of 240 µm and an SNR of 34 for this acquisition setting [
18]. While the system is capable of capturing a voxel size of 40 µm, the exposure and reconstruction times would have to be increased to non-clinically relevant timescales if scanning the entire field of view. For selected scans, 3D visualization models were created using the “Volume Rendering” tool in 3D Slicer (v4,
https://www.slicer.org) [
19].
Pathological diagnosis
After micro-CT scanning, specimens were returned to the pathologists’ assistants, who performed standard-of-care gross analysis to determine both representative sections of lesional tissue as well as all areas of possible lesion involvement with the specimen margin. These tissue sections underwent standard histological processing to create hematoxylin and eosin (H&E) stained slides, which were diagnosed by an experienced histopathologist (W.A.W.). From these slides, distances from lesion involvement to each margin were measured. Tabulated diagnoses and distances associated with each margin of each specimen are included in Supplemental File 1.
Micro-CT image interpretation
After scans were acquired, an experienced breast radiologist (R.A.Z.) retrospectively analyzed the scans, which were sequentially (scan-mode) presented in two orthogonal views. Features such as speculated masses, distorted tissue architecture, and clustered micro-calcifications were evaluated to yield a binary reading of positive or negative margins. The radiologist was blinded to the final pathological diagnosis as well as the pathological diagnosis of the pre-operative needle core biopsy. For four specimens, micro-CT scans were compared retrospectively to de-identified single-projection specimen mammograms to view qualitatively the effect of micro-CT reconstruction on overlapping structures in the projection images.
Statistical methods
All statistical analyses were performed using MATLAB (v2016a, Mathworks, Natick, MA). For reader performance metrics, confidence intervals were calculated using a binomial distribution. A one-sided Fisher’s exact test was used to calculate statistical significance between observed involvement in the different surface planes.
Discussion
Positive surgical margins after BCS remain a notorious clinical challenge. Wire localization and specimen mammography serve as the primary forms of intra-operative guidance but have led to re-excision rates ranging from 0 to 70% [
10]. Microscopic point assessment techniques, such as frozen-section pathology and touch prep cytology, have been used for directed BCS guidance with apparent diagnostic accuracy, but wide-spread adoption has been limited due to increasing labor costs of microscopic sampling over a macroscopic specimen [
21]. Accordingly, an advantage of micro-CT demonstrated in this study is its ability to image the entire specimen in 3D, thereby enabling analyses of the complete specimen.
Although modest sensitivity and specificity (50%, 2/4, 67%, 16/24) were reported in this study, significantly more pathologically false positives (8/28) occurred than pathologically false negatives (2/28) (
p = 0.04) for malignant lesions. While excising unnecessary tissue is less undesirable, it may be less harmful than a second surgical procedure. This notion is highlighted by the proposed practice of undirected whole margin cavity shaving on all BCS procedures, which has recently been investigated [
22]. Furthermore, single-projection specimen mammography has reported sensitivity and specificity of 36% and 71% [
7], whereas standard-of-care two projection specimen mammography has yielded sensitivities and specificities of 54.6% and 87.8% [
23], and more recently 58.5% and 91.8% [
24]. While the preliminary micro-CT results presented here suggest improved sensitivity and comparable specificity to single-projection mammography, and comparable sensitivity but inferior specificity to two projection mammography, a prospective, fully powered, head to head study of micro-CT against standard-of-care two projection specimen mammography is needed before the performances of the two modalities can be accurately compared. Additionally, the reading of micro-CT specimen scans was not optimized and no training was provided, the latter would benefit from an atlas of micro-CT scans of BCS specimens but none currently exists. The data presented here could serve as the starting point for a training set for the presentation of various pathologies in micro-CT. Nevertheless, reading individual cross-sectional planes of specimens in the projection views micro-CT appears to offer valuable information on lesion location.
This inceptive study provides initial evidence suggesting that micro-CT specimen scanning offers potential diagnostic value, which could be quantified in a larger prospective study. While this study demonstrated the ability to detect in-situ disease from the analysis of clustered micro-calcifications, a larger sample size is needed to more reliably quantify the accuracy of detecting invasive versus in-situ disease. A notable limitation of micro-CT was its lack of contrast between tumor-glandular tissue, which resulted in poor diagnosis of tumor-glandular interfaces. While a lower kVp could increase image contrast, this would require a longer exposure time to maintain photon counts and SNR. Alternatively, dual energy contrast enhancement could be envisioned to increase contrast, but this would require two scans at different energies as well as the injection of a contrast agent [
25]. Another limitation was the lack of radiologic training available prior to the study. In future, prospective studies, the data presented in this manuscript could be used as a radiologic atlas and training data set, which could potentially improve reader performance. Another limiting factor was the presence of metal streaking artifacts arising in reconstructions of specimens containing localization wires. In Tang et al., this difficulty was mitigated by first performing a single-projection image to confirm the wire and surgical clip were excised, then removing the wire from the specimen, and performing the full micro-CT scan [
14]. However, this approach eliminates the wire and clip which are used regularly as lesion landmarks during subsequent gross analysis in Pathology. An alternative solution would be to implement a metal artifact reduction algorithm, such as segmenting the metal, manually setting its known attenuation coefficient, and performing an iterative reconstruction [
26,
27]. Finally, it should be noted that the gold-standard histopathological margin analysis, which samples only a subset of the margin chosen through gross dissection and evaluation, has shown to only weakly correlate with the actual presence of disease in the lumpectomy cavity, with reported accuracies as low as 64.9% and up to 20% of pathologically negative margins presenting residual disease [
28,
29]. This raises the possibility that micro-CT may be able to detect disease invisible during gross dissection and subsequent pathological analysis, and motives the need for more spatially exact correlations between micro-CT readings and pathological confirmation in future studies.
In summary, the results reported here demonstrate the feasibility of using micro-CT to provide full volume 3D microscopic reconstructions of lumpectomy specimens that highlight spatially-resolved features not evident in conventional 2D projection radiography. Furthermore, the study offers an initial measure of the correlation between micro-CT analysis of whole BCS specimens and current gold-standard histopathology. Micro-CT should continue to be investigated as a potential BCS guidance tool because of its rapid acquisition and volume rendering capabilities.