Artificial intelligence (AI) for breast cancer screening: BreastScreen population-based cohort study of cancer detection.

BACKGROUND: Artificial intelligence (AI) has been proposed to reduce false-positive screens, increase cancer detection rates (CDRs), and address resourcing challenges faced by breast screening programs. We compared the accuracy of AI versus radiologists in real-world population breast cancer screening, and estimated potential impacts on CDR, recall and workload for simulated AI-radiologist reading. METHODS: External validation of a commercially-available AI algorithm in a retrospective cohort of 108,970 consecutive mammograms from a population-based screening program, with ascertained outcomes (including interval cancers by registry linkage). Area under the ROC curve (AUC), sensitivity and specificity for AI were compared with radiologists who interpreted the screens in practice. CDR and recall were estimated for simulated AI-radiologist reading (with arbitration) and compared with program metrics. FINDINGS: The AUC for AI was 0.83 compared with 0.93 for radiologists. At a prospective threshold, sensitivity for AI (0.67; 95% CI: 0.64-0.70) was comparable to radiologists (0.68; 95% CI: 0.66-0.71) with lower specificity (0.81 [95% CI: 0.81-0.81] versus 0.97 [95% CI: 0.97-0.97]). Recall rate for AI-radiologist reading (3.14%) was significantly lower than for the BSWA program (3.38%) (-0.25%; 95% CI: -0.31 to -0.18; P < 0.001). CDR was also lower (6.37 versus 6.97 per 1000) (-0.61; 95% CI: -0.77 to -0.44; P < 0.001); however, AI detected interval cancers that were not found by radiologists (0.72 per 1000; 95% CI: 0.57-0.90). AI-radiologist reading increased arbitration but decreased overall screen-reading volume by 41.4% (95% CI: 41.2-41.6). INTERPRETATION: Replacement of one radiologist by AI (with arbitration) resulted in lower recall and overall screen-reading volume. There was a small reduction in CDR for AI-radiologist reading. AI detected interval cases that were not identified by radiologists, suggesting potentially higher CDR if radiologists were unblinded to AI findings. These results indicate AI's potential role as a screen-reader of mammograms, but prospective trials are required to determine whether CDR could improve if AI detection was actioned in double-reading with arbitration. FUNDING: National Breast Cancer Foundation (NBCF), National Health and Medical Research Council (NHMRC).

Localisation accuracy with iodine-125 seed versus wire guidance for breast cancer surgery.

INTRODUCTION: Impalpable breast lesions generally require image-guided localisation for breast-conserving surgery. A standard technique is to place a hook wire (HW) within the lesion. Radioguided occult lesion localisation using iodine seeds (ROLLIS) involves inserting a 4.5 mm iodine-125 seed (seed) into the lesion. We hypothesised that a seed could be more precisely positioned in relation to the lesion than a HW and that this may be associated with a lower re-excision rate. METHODS: Retrospective review of consecutive participant data from three ROLLIS RCT (ACTRN12613000655741) sites. Participants underwent preoperative lesion localisation (PLL) with seed or HW between September 2013 and December 2017. Lesion and procedural characteristics were recorded. Distances between (1) any part of the seed or thickened segment of the HW ('TSHW') and the lesion/clip ('distance to device' DTD) and (2) centre of the TSHW/seed and centre of the lesion/clip (device centre to target centre 'DCTC') were measured on immediate postinsertion mammograms. Pathological margin involvement and re-excision rates were compared. RESULTS: A total of 390 lesions (190 ROLLIS and 200 HWL) were analysed. Lesion characteristics and guidance modality used were similar between groups. Ultrasound-guided DTD and DCTC for seed were smaller than for HW (77.1% and 60.6%, respectively, P-value < 0.001). Stereotactic-guided DCTC for seeds was 41.6% smaller than for HW (P-value = 0.001). No statistically significant difference in the re-excision rates was found. CONCLUSION: Iodine-125 seeds can be more precisely positioned for preoperative lesion localisation than HW, however, no statistically significant difference in re-excision rates was detected.

Artifacts in contrast-enhanced mammography: are there differences between vendors?

PURPOSE: Contrast-Enhanced Mammography (CEM) produces a dual-energy subtracted (DES) image that demonstrates iodine uptake (neovascularity) in breast tissue. We aim to review a range of artifacts on DES images produced using equipment from two different vendors and compare their incidence and subjective severity. METHODS: We retrospectively reviewed CEM studies performed between September 2013 and March 2017 using GE Senographe Essential (n = 100) and Hologic Selenia Dimensions (n = 100) equipment. Artifacts were categorized and graded in severity by a subspecialist breast radiologist and one of two medical imaging technologists in consensus. The incidence of artifacts between vendors was compared by calculating the relative risk, and the severity gradings were compared using a Wilcoxon rank-sum test. RESULTS: Elephant rind, corrugations and the black line on chest wall artifact were seen exclusively in Hologic images. Artifacts such as cloudy fat, negative rim around lesion and white line on pectoral muscle were seen in significantly more Hologic images (p < 0.05) whilst halo, ripple, skin line enhancement, black line on pectoral muscle, bright pectorals, chest wall high-lighting and air gap were seen in significantly more GE images (p < 0.05). The severity gradings for cloudy fat had a significantly higher mean rank in Hologic images (p < 0.001) whilst halo and ripple artifacts had a significantly higher mean rank in GE images (p < 0.001 and p = 0.028 respectively). CONCLUSION: The type, incidence and subjective severity of CEM-specific artifacts differ between vendors. Further research is needed, but differences in algorithms used to produce the DE image are postulated to be a significant contributor.

A comparison of stereotactic and tomosynthesis-guided localisation of impalpable breast lesions.

INTRODUCTION: Impalpable breast cancers require precise pre-operative lesion localisation to minimise re-excision rates. Conventional techniques include hookwire insertion using stereotactic guidance. Newer techniques include the use of tomosynthesis guidance and the use of iodine-125 seeds. This study compares the accuracy of lesion localisation with hookwire or seed insertion using prone stereotactic or upright tomosynthesis guidance. METHODS: This registered quality improvement activity did not require formal ethics approval. The post-localisation images for 116 lesions were reviewed. The distance from the lesion or breast biopsy marker to the hookwire or seed was measured on post-insertion mammograms. The relative placement accuracy of hookwire or seed using prone stereotactic or upright tomosynthesis guidance was compared. A lesion to seed or wire distance > 10 mm was considered technically unsatisfactory. RESULTS: 94.8% of the seeds and wires inserted via prone stereotactic guidance were accurately placed, compared with 89.6% of those inserted via upright tomosynthesis. There were twice as many technically unsatisfactory insertions under upright tomosynthesis guidance. The majority of the unsatisfactory insertions using upright tomosynthesis occurred when the lesion was at or below the level of the nipple and the insertion was performed craniocaudally. CONCLUSION: The degree of accuracy of pre-operative localisation of impalpable breast lesions is significantly higher with the use of prone stereotactic rather than upright tomosynthesis guidance. This was most evident with the placement of I-125 seeds, and in cases where the target lesion was located below the level of the nipple.

Radiographer technique: Does it contribute to the question of clip migration?

INTRODUCTION: Marker clips are commonly deployed at the site of a percutaneous breast biopsy. Studies have shown that displacement of the clip from the site of deployment is not uncommon. The objective of this study was to determine how much 'migration' could be seen with fixed structures within the breast tissue across three consecutive annual screening examinations, and therefore attempt to quantify how much of the reported clip migration could be due to radiographer technique. METHODS: Large, easily identified benign calcifications were measured by two investigators across three consecutive cycles of screening mammography. The position of the calcifications on the two standard mammographic views was measured in two planes. Other variables recorded included breast size and density, compression force used, and location of the benign calcifications within the breast. RESULTS: In 38% of cases, benign breast calcifications showed a mimicked movement of >15 mm in at least one plane. This was greatest in large breasts, those where fibroglandular tissue occupied less than 50% of the breast volume, and in the upper outer quadrant of the breast where mimicked movement >10 mm was noted in up to 90% of the larger breasts. CONCLUSION: Fixed immobile objects in the breast can appear to move a distance of >15 mm in up to 30% of cases. Clinically, some of what has previously been called marker 'migration' may be spurious and accounted for by differences in radiographic positioning techniques.