Bo La Yun

Background/Objectives: Accurate visual categorization of small enhancing lesions on contrast-enhanced breast MRI remains challenging. The purpose of this study was to develop a deep learning (DL) model for detecting small (≤2 cm) invasive breast cancers using multi-institutional breast MRI data. Methods: This retrospective study included a total of 1721 women (mean age: 54 years; range, 21–87 years) with T1-stage invasive breast cancer across five hospitals. All patients underwent dynamic contrast-enhanced breast MRI before surgery between 2010 and 2020. Enhancing breast masses on early postcontrast T1-weighted images (14,917 labeled cancer images and 1443 labeled benign images) were used as the ground truth. In this multicenter cancer-enriched cohort, a DL model was developed to distinguish cancerous lesions from noncancerous lesions. MR images were independently reviewed by three breast imaging radiologists to detect subcentimeter (≤1 cm) cancers, and the readers were informed that each examination contained a single cancer. The performance of the model was evaluated using the area under the precision–recall curve (AUPRC), sensitivity, precision, and F1 score. Results: The DL model achieved an AUPRC of 0.42 in detecting small (≤2 cm) invasive breast cancers presenting as enhancing masses on MRI (sensitivity, 83.2%; precision, 33.2%; F1 score, 0.47). For subcentimeter (≤1 cm) cancers, the detection performance of the model was lower (sensitivity, 75.3%; precision, 20.8%; F1 score, 0.33) than that of the radiologists (mean sensitivity, 89.5%; mean precision, 72.9%; mean F1 score, 0.80). When the radiologists used the DL model, their mean precision in detecting subcentimeter invasive breast cancers improved from 72.9% to 83.2%, with no significant change in sensitivity (89.5% vs. 86.8%, p > 0.05). Conclusions: The DL model demonstrated potential as an assistive tool by improving radiologists’ precision for detecting small invasive breast cancers on contrast-enhanced breast MRI in this enriched reader-study setting, although it did not significantly improve their sensitivity.

Background/Objectives: We compared the visibility of breast cancer using the newly developed standing automated breast ultrasound system (MammouS-N) and digital breast tomosynthesis (DBT), and identified factors influencing lesion visibility. Methods: We prospectively enrolled 100 women (mean age: 51.6 years; range: 26–76 years) who were diagnosed with breast cancer and were scheduled to undergo DBT between January and July 2024. They underwent DBT and an ultrasound on the same day. Two radiologists evaluated the visibility scores (0–5) of lesions corresponding to biopsy-confirmed breast cancers identified using magnetic resonance imaging. The Wilcoxon signed-rank test was used to compare the visibility scores of cancers identified on DBT and/or MammouS-N images. Results: Among the 100 women, invasive ductal carcinoma was the most common malignancy (73%). DBT findings included negative findings (7%), masses (46%), masses with calcification (29%), calcifications only (15%), and architectural distortions (3%). On MammouS-N ultrasound, most lesions were classified as masses (93%), whereas 7% were non-mass lesions. For Reviewer 1, MammouS-N demonstrated significantly higher visibility scores (higher scores: 26 on MammouS-N, seven on DBT; equal scores: 67, z = −3.234, p = 0.001). For Reviewer 2, the two modalities showed no significant difference in visibility (higher scores: 27 on MammouS-N, 28 on DBT, equal scores: 45, z = −0.040, p = 0.968). Noncalcified lesions that were obscured on DBT were better visualized on MammouS-N (p < 0.001) by both reviewers. Conclusions: MammouS-N holds promise as an imaging modality complementary to DBT in women with dense breast tissue, particularly for non-calcified lesion detection.

To assess the diagnostic performance and clinical usefulness of deep learning-based computer-aided detection (AI-CAD) for automated breast ultrasound (ABUS) across radiologists with varying ABUS experience. This retrospective study included 114 women (228 breasts) who underwent ABUS in 2019. Three radiologists interpreted images with and without AI-CAD. We evaluated sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and area under the curve (AUC), reading time and interobserver agreement in Breast Imaging Reporting & Data System (BI-RADS) categorization and biopsy recommendations. Among 114 women (50.9 ± 10.8 years), 28 were diagnosed with breast cancer. The following performance metrics improved significant with AI-CAD: Reader 1 (least experienced of ABUS; 2 years of ABUS experience), AUC, 0.837 to 0.947 (p = 0.009), and NPV, 95.8% to 98.4% (p = 0.022); Reader 2 (7 years of experience), PPV, 50.0% to 59.5% (p = 0.042); Reader 3 (8 years of experience), PPV, 55.6% to 66.7% (p = 0.034). Reader 1 with AI-CAD achieved a performance comparable or higher than those of more experienced readers without AI. Specifically: compared with Reader 2, specificity (93.5% vs. 88.0%), PPV (65.8% vs. 50.0%), and accuracy (93.0% vs. 87.7%) were higher. Although Reader 3 originally demonstrated higher NPV (98.4% vs. 95.8%) and AUC (0.954 vs. 0.837) without CAD, these differences were no longer significant when Reader 1 used AI-CAD. Across all readers, AI-CAD reduced the mean reading time by an average of 25 s (p < 0.001). Inter-observer agreement after AI-CAD use (BI-RADS κ: 0.279 → 0.363; biopsy recommendation κ: 0.666 → 0.736) showed no statistically significant difference. AI-CAD enhanced diagnostic performance and reading efficiency in ABUS interpretation, demonstrating the most pronounced improvement for the less experienced reader.

Background/Objectives: To identify preoperative MRI features using computer-assisted diagnosis (CAD) that are associated with invasive disease-free survival (IDFS) and distant metastasis-free survival (DDFS) in patients with primarily operable triple-negative breast cancer (TNBC). Methods: This retrospective study was approved by the institutional review board with informed consent was waived. Between January 2012 and December 2014, 74 consecutive women with primary TNBC (mean age, 51 years; range, 29–77 years) who underwent preoperative MRI were included and followed until August 2021. Dynamic contrast-enhanced and T2-weighted images were obtained using 3T scanners. Peritumoral edema and central necrosis were evaluated retrospectively. CAD was used to extract 3D diameters, angiovolume, and kinetic parameters, and kinetic heterogeneity was calculated. Cox proportional hazards models were used to assess associations between MRI features and IDFS and DDFS, adjusting for clinicopathologic factors. Results: During a median follow-up of 80.9 months, 12 patients developed invasive disease, and 8 developed distant metastasis. In multivariable analysis, peak enhancement (hazard ratio [HR], 1.40; 95% confidence interval [CI], 1.06–1.84; p = 0.019) and angiovolume (HR, 2.86; 95% CI, 1.26–6.47; p = 0.012) were independently associated with IDFS, whereas angiovolume (HR, 2.47; 95% CI: 1.28–4.78; p = 0.007) was independently associated with DDFS. Conclusions: Preoperative CAD-derived MRI features, particularly peak enhancement and angiovolume, were associated with IDFS in TNBC patients whereas angiovolume alone was associated with DDFS.

Kim YS, et al. Korean J Radiol. 2024 Jun;25(6):511-517. https://doi.org/10.3348/kjr.2023.1189

Kim SM, et al. Korean J Radiol. 2024 Feb;25(2):146-156. https://doi.org/10.3348/kjr.2023.0100

Kim YS, et al. Korean J Radiol. 2024 Jan;25(1):11-23. https://doi.org/10.3348/kjr.2023.0528

A substantial percentage of women die from cardiovascular disease (CVD). Computed tomography (CT) scan helps predict/monitor CVD-related diseases. However, previous studies found relation to assessing the risk of CVD by estimating severity of breast arterial calcification (BACs). In many countries, women undergo breast cancer screening using mammograms regularly, so it is easier to access vast datasets compared to CT-scan without additional cost and absorbing radiation. Earlier, measuring BACs severity was largely depends on the performance of segmentation of calcified vessel. Nevertheless, dataset maintenance for segmentation is expensive and tedious. As pixel-wise manual ground-truth (GT) marking can only be performed by multiple experienced experts. First time ever, to the best of our knowledge, a direct ordinal measurement of BACs severity is proposed without prior segmentation. Therefore, due to the complexity of calcified vessels, cascade classification performs better than end-to-end classification. In this proposed method, first, we imply Frangi filter to deduct background automatically without any prior ground-truth marking. Then, unlike other methods, interpolation is implied to shrink cropped images rather than splitting images. Also, left view images have flipped to reduce the complexity of different directions of breasts. In addition, among various residual network (ResNet) architecture, proposed method customized ResNet50 to adjust large image size and vessel complexity. The proposed method reduces cost, time, and tedious job of researchers as this does not require segmenting calcified vessels and provides efficient prediction of BACs severity which is helpful to assess the risk of CVD in long run.

Jung JJ, et al. J Breast Cancer. 2023 Mar;26:e14. https://doi.org/10.4048/jbc.2023.26.e14

Choi Y, et al. Korean J Radiol. 2022 Sep;23(9):866-877. https://doi.org/10.3348/kjr.2021.0850