The role of tomosynthesis in intraoperative specimen evaluation

Analysis of Specimen Mammography with Artificial Intelligence to Predict Margin Status

BACKGROUND

Intraoperative specimen mammography is a valuable tool in breast cancer surgery, providing immediate assessment of margins for a resected tumor. However, the accuracy of specimen mammography in detecting microscopic margin positivity is low. We sought to develop an artificial intelligence model to predict the pathologic margin status of resected breast tumors using specimen mammography.

METHODS

A dataset of specimen mammography images matched with pathologic margin status was collected from our institution from 2017 to 2020. The dataset was randomly split into training, validation, and test sets. Specimen mammography models pretrained on radiologic images were developed and compared with models pretrained on nonmedical images. Model performance was assessed using sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC).

RESULTS

The dataset included 821 images, and 53% had positive margins. For three out of four model architectures tested, models pretrained on radiologic images outperformed nonmedical models. The highest performing model, InceptionV3, showed sensitivity of 84%, specificity of 42%, and AUROC of 0.71. Model performance was better among patients with invasive cancers, less dense breasts, and non-white race.

CONCLUSIONS

This study developed and internally validated artificial intelligence models that predict pathologic margins status for partial mastectomy from specimen mammograms. The models' accuracy compares favorably with published literature on surgeon and radiologist interpretation of specimen mammography. With further development, these models could more precisely guide the extent of resection, potentially improving cosmesis and reducing reoperations.

Computer Vision Analysis of Specimen Mammography to Predict Margin Status

Intra-operative specimen mammography is a valuable tool in breast cancer surgery, providing immediate assessment of margins for a resected tumor. However, the accuracy of specimen mammography in detecting microscopic margin positivity is low. We sought to develop a deep learning-based model to predict the pathologic margin status of resected breast tumors using specimen mammography. A dataset of specimen mammography images matched with pathology reports describing margin status was collected. Models pre-trained on radiologic images were developed and compared with models pre-trained on non-medical images. Model performance was assessed using sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). The dataset included 821 images and 53% had positive margins. For three out of four model architectures tested, models pre-trained on radiologic images outperformed domain-agnostic models. The highest performing model, InceptionV3, showed a sensitivity of 84%, a specificity of 42%, and AUROC of 0.71. These results compare favorably with the published literature on surgeon and radiologist interpretation of specimen mammography. With further development, these models could assist clinicians with identifying positive margins intra-operatively and decrease the rate of positive margins and re-operation in breast-conserving surgery.

Diagnostic performance of tomosynthesis, digital mammography and a dedicated digital specimen radiography system versus pathological assessment of excised breast lesions

BACKGROUND

The aim of the study was to compare the performance of full-field digital mammography (FFDM), digital breast tomosynthesis and a dedicated digital specimen radiography system (SRS) in consecutive patients, and to compare the margin status of resected lesions versus pathological assessment.

PATIENTS AND METHODS

Resected tissue specimens from consecutive patients who underwent intraoperative breast specimen assessment following wide local excision or oncoplastic breast conservative surgery were examined by FFDM, tomosynthesis and SRS. Two independent observers retrospectively evaluated the visibility of lesions, size, margins, spiculations, calcifications and diagnostic certainty, and chose the best performing method in a blinded manner.

RESULTS

We evaluated 216 specimens from 204 patients. All target malignant lesions were removed with no tumouron-ink. One papilloma had positive microscopic margins and one patient underwent reoperation owing to extensive in situ components. There were no significant differences in measured lesion size among the three methods. However, tomosynthesis was the most accurate modality when compared with the final pathological report. Both observers reported that tomosynthesis had significantly better lesion visibility than SRS and FFDM, which translated into a significantly greater diagnostic certainty. Tomosynthesis was superior to the other two methods in identifying spiculations and calcifications. Both observers reported that tomosynthesis was the best performing method in 76.9% of cases. The interobserver reproducibilities of lesion visibility and diagnostic certainty were high for all three methods.

CONCLUSIONS

Tomosynthesis was superior to SRS and FFDM for detecting and evaluating the target lesions, spiculations and calcifications, and was therefore more reliable for assessing complete excision of breast lesions.

Diagnostic Performance of Adjunctive Imaging Modalities Compared to Mammography Alone in Women with Non-Dense and Dense Breasts: A Systematic Review and Meta-Analysis

PURPOSE

To compare the diagnostic performance of mammography (MG) alone versus MG combined with adjunctive imaging modalities, including handheld ultrasound (HHUS), automated breast ultrasound (ABUS), digital breast tomosynthesis (DBT), contrast-enhanced mammography (CEM), and magnetic resonance imaging (MRI) in women with non-dense and dense breasts.

PATIENTS AND METHODS

Medline, Embase, PubMed, CINAHL, Scopus, and the Web of Science databases were searched up to October 2019. Quality assessment was performed using QUADAS-2. RevMan 5.3 was used to conduct a meta-analysis of the studies.

RESULTS

In dense breasts, adding adjunctive modalities significantly increased cancer detection rates (CDRs): HHUS (relative risk [RR] = 1.49; 95% confidence interval [CI], 1.19-1.86; P = .0005); ABUS (RR = 1.44; 95% CI, 1.16-1.78; P = .0008); DBT (RR = 1.38; 95% CI, 1.14-1.67; P = .001); CEM (RR = 1.37; 95% CI, 1.12-1.69; P = .003); and MRI (RR = 2.16; 95% CI, 1.81-2.58; P < .00001). The recall rate was significantly increased by HHUS (RR = 2.03; 95% CI, 1.89-2.17; P < .00001), ABUS (RR = 1.90; 95% CI, 1.81-1.99; P < .00001), and MRI (RR = 2.71; 95% CI, 1.73-4.25; P < .0001), but not by DBT (RR = 1.14; 95% CI, 0.95-1.36; P = .15). In non-dense breasts, HHUS and MRI showed significant increases in CDRs but not DBT: HHUS (RR = 1.14; 95% CI, 1.01-1.29; P = .04); MRI (RR = 1.78; 95% CI, 1.14-2.77; P = .01); and DBT (RR = 1.09; 95% CI, 1.13-1.75; P = .08). The recall rate was also significantly increased by HHUS (RR = 1.43; 95% CI, 1.28-1.59; P < .00001) and MRI (RR = 3.01; 95% CI, 1.68-5.39; P = .0002), whereas DBT showed a non-significant reduction (RR = 0.83; 95% CI, 0.65-1.05; P = .12).

CONCLUSION

Adding adjunctive modalities to MG increases CDRs in women with dense and non-dense breasts. Ultrasound and MRI increase recall rates across all breast densities; however, MRI results in higher values for both CDRs and recall rates.

Intraoperative digital breast tomosynthesis using a dedicated device is more accurate than standard intraoperative mammography for identifying positive margins

AIM

To compare a standard intra-operative mammography (IM) device with digital breast tomosynthesis using a dedicated device (Mozart system) in the evaluation of surgical margins at first excision.

MATERIALS AND METHODS

The study received institutional review board approval and written informed consent was obtained from participants. From January 2018 to December 2018, a prospective analysis of the images of IM device and intra-operative digital breast tomosynthesis with a dedicated device (Mozart system) in n=89 breast cancer patients (average patients age: 58 years, age range: 35-76 years) was undertaken. Images were evaluated by two expert breast radiologists independently of each other and blinded to each other's interpretation, who indicated the positive cases requiring surgical re-excision intra-operatively.

RESULTS

Mean cancer size was 12.5±4.5 mm. Radiological signs of the lesions were microcalcifications (n=71), nodules (n=10), and architectural distortions (n=8). A total of 20/89 (17%) patients underwent intra-operative re-excision for positive margins. Intra-operative digital breast tomosynthesis with a dedicated device and IM showed discrepancies in 15/89 cases (17%). Mozart system results informed the necessity to perform a re-excision (n=15). Overall, receiver operating characteristic (ROC) curve analysis showed and area under the ROC curve (AUC) of 0.82 for the Mozart system versus 0.65 for IM. ROC analysis of radiological findings with microcalcifications showed an AUC of 0.92 for the Mozart system versus 0.74 for IM, whereas AUC in cases with no microcalcifications were 0.87 and 0.75, respectively.

CONCLUSION

Intra-operative digital breast tomosynthesis with a dedicated device provides more information (better accuracy) than IM and facilitated a reduction in re-excision rates.

Breast lesion size assessment in mastectomy specimens: Correlation of cone-beam breast-CT, digital breast tomosynthesis and full-field digital mammography with histopathology

To compare the accuracy of breast lesion size measurement of cone-beam breast-CT (CBBCT), digital breast tomosynthesis (DBT) and full-field digital mammography (FFDM).Patients scheduled for mastectomy due to at least 1 malignant breast lesion were included. Mastectomy specimens were examined by CBBCT, DBT, FFDM, and histopathology.A total of 94 lesions (40 patients) were included. Histopathological analyses revealed 47 malignant, 6 high-risk, and 41 benign lesions. Mean histopathological lesion size was 20.8 mm (range 2-100). Mean absolute size deviation from histopathology was largest for FFDM (5.3 ± 6.7 mm) and smallest for CBBCT 50 mA, high-resolution mode (4.3 ± 6.7 mm). Differences between imaging modalities did not reach statistical significance (P = .85).All imaging methods tend to overestimate breast lesion size compared to histopathological gold standard. No significant differences were found regarding size measurements, although in tendency CBBCT showed better lesion detection and cT classification over FFDM.

Digital Breast Tomosynthesis: What Have We Learned?

Digital breast tomosynthesis (DBT) is increasingly recognized as a superior breast imaging technology compared with 2D digital mammography (DM) alone. Accumulating data confirm increased sensitivity and specificity in the screening setting, resulting in higher cancer detection rates and lower abnormal interpretation (recall) rates. In the diagnostic environment, DBT simplifies the diagnostic work-up and improves diagnostic accuracy. Initial concern about increased radiation exposure resulting from the DBT acquisition added onto a 2D mammogram has been largely alleviated by the development of synthesized 2D mammography (SM). Continued research is underway to reduce artifacts associated with SM, and improve its comparability to DM. Breast cancers detected with DBT are most often small invasive carcinomas with a preponderance for grade 1 histology and luminal A molecular characteristics. Recent data suggest that higher-grade cancers are also more often node negative when detected with DBT. A meta-analysis of early single-institution studies of the effect of DBT on interval cancers has shown a modest decrease when multiple data sets are combined. Because of the greater conspicuity of lesions on DBT imaging, detection of subtle architectural distortion is increased. Such findings include both spiculated invasive carcinomas and benign etiologies such as radial scars. The diagnostic evaluation of architectural distortion seen only with DBT can pose a challenge. When no sonographic correlate can be identified, DBT-guided biopsy and/or localization capability is essential. Initial experience with DBT-guided procedures suggests that DBT biopsy equipment may improve the efficiency of percutaneous breast biopsy with less radiation.