The clinical problem the researchers set out to solve is one that radiologists know all too well. Breast lesions classified as BI-RADS 4 or 5 — meaning suspicious enough to warrant biopsy — cover a wide spectrum of malignancy risk, and their ultrasound appearances overlap maddeningly. A benign lesion can look irregular in shape with indistinct margins and a dark, hypoechoic texture; so can a cancer. Posterior acoustic attenuation, a shadowing effect once considered a red flag, appears in both. This morphological ambiguity produces high inter-observer variability, particularly among less experienced practitioners. Ultrasound diagnosis is also highly operator-dependent, shaped by scanning skill, image acquisition standards, and subjective judgment. The stakes are high on both sides: a missed malignancy delays treatment, while an unnecessary core-needle biopsy or surgery exposes a patient to risk, anxiety, and cost for what turns out to be a harmless finding.

The challenge is compounded by breast tissue density. Asian women, on average, have denser breast tissue than other populations, which degrades the performance of mammography, the traditional screening mainstay. In China, clinical guidelines therefore pair mammography with ultrasound as the primary and auxiliary imaging methods for breast cancer detection. Ultrasound is convenient, radiation-free, and particularly effective in dense tissue — but its accuracy in distinguishing benign from malignant BI-RADS 4–5 lesions has remained stubbornly limited. Previous computational efforts largely focused on static images, requiring the sonographer or an algorithm to pick a single representative frame from a live scan. That choice discards a wealth of information: the way a lesion’s boundaries shift as the probe moves, how its internal echoes evolve, and the temporal context that an experienced human eye implicitly absorbs during a real-time examination.

TGMIL was designed specifically to exploit that discarded information. The architecture treats each ultrasound video as a bag of frames in a machine learning framework known as multi-instance learning, where the overall diagnosis is known but the diagnostic value of each individual frame is not. Each frame is first encoded by a VGG-13 convolutional backbone, generating a sequence of frame-level features. Crucially, the model also encodes the temporal order of frames through a linear layer, concatenating this positional information with the visual features so the network can learn from how a lesion appears, evolves, and disappears across the scan. An attention module then adaptively assigns higher weights to the most diagnostically informative frames and aggregates them into a single video-level representation. In effect, TGMIL learns which moments of the scan matter — suppressing redundant frames and emphasizing clinically relevant temporal segments — rather than treating every image equally.

The study enrolled 532 breast lesions categorized as BI-RADS 4–5 across four hospitals between November 2019 and January 2023. Of these, 334 lesions (61 benign and 273 malignant) formed the training and validation sets, while 198 lesions (50 benign and 148 malignant) were reserved as an independent test set. All videos were acquired with high-frequency linear probes, scanned continuously from the first appearance of each lesion to its complete disappearance, and converted to DICOM format. Preprocessing was rigorous: videos were converted to a lossless format, black borders and text overlays were stripped, a K-means algorithm extracted 15 key frames per video, and pixel values were standardized. Every diagnosis was confirmed by pathology from surgery or ultrasound-guided biopsy, and all examinations from the same patient were kept within a single dataset partition to prevent data leakage.

On the external test set, TGMIL achieved an area under the receiver operating characteristic curve of 0.824, decisively outperforming mainstream video classification architectures including 3D-CNN (AUC 0.665), CSN (0.613), and R(2+1)D-50 (0.678). It also beat static image-based classifiers: ResNet50, VGG16, the transformer-based ViT B/32 (which managed only 0.656), and even BU-Mamba, a recently proposed static ultrasound model. BU-Mamba achieved remarkable sensitivity of 0.988 but suffered from low precision of 0.752, indicating a tendency toward over-diagnosis. TGMIL struck a better balance, pairing high sensitivity of 0.966 with precision of 0.836 — meaning it reliably caught malignancies without flooding the clinic with false alarms. Ablation tests confirmed the value of the attention mechanism: simple mean pooling and max pooling baselines reached AUCs of only 0.752 and 0.770 respectively, showing that adaptively identifying informative frames, rather than averaging or maximizing across them, is what drives the performance gain.

The most clinically compelling results came from the reader study. Nine ultrasonographers — three juniors with under three years of experience, three attendings with four to nine years, and three seniors with a decade or more — independently reviewed the 198 test lesions, first without AI assistance and then, four weeks later and in a reshuffled case order, with TGMIL’s real-time benign-versus-malignant probability estimates. Unaided, the model itself achieved a patient-based AUC of 0.845, significantly outperforming both junior sonographers (0.682) and attendings (0.802). With TGMIL’s help, junior accuracy jumped from 66.7 to 89.2 percent and sensitivity surged from 67.4 to 94.2 percent, while attending accuracy rose from 80.8 to 90.0 percent. Across all readers, pooled accuracy improved from 77.5 to 95.0 percent, and the pooled AUC climbed from 0.871 to 0.964. Specificity for the whole group rose from 88.2 to 97.1 percent — a direct strike against unnecessary biopsies.

Interestingly, the most experienced readers gained the least. Senior ultrasonographers, who already achieved an AUC of 0.903 unaided, showed no statistically significant improvement on any metric with AI assistance, and their AUC dipped slightly from 0.898 to 0.885. The authors offer a candid explanation: because TGMIL’s standalone performance (0.824) falls below that of seasoned experts, the model may add little information beyond what veterans already extract from morphology, Doppler signals, and clinical intuition. Worse, when model predictions conflict with expert judgment, uncertain outputs can introduce cognitive interference or automation bias, occasionally nudging a correct decision astray. The study’s failure analysis reinforced this caution: some benign lesions displayed pseudo-malignant features such as spiculated margins that fooled the model into false positives, while some cancers with weak or atypical vascular signals escaped detection as false negatives.

The practical implications are nonetheless considerable. TGMIL was trained on a single NVIDIA RTX 3090 GPU — no supercomputer required — and once trained, inference on ultrasound video is computationally light enough for a local workstation, meaning the system could be integrated directly into the ultrasound workflow. The model also showed strong probability calibration (Brier score 0.071) and robust performance across all four external test centers, with AUCs ranging up to 0.886, suggesting it generalizes across different scanners, hospitals, and patient populations. For regions facing shortages of senior breast imaging specialists, an AI assistant that lifts junior sonographers to near-expert accuracy could meaningfully expand access to reliable diagnosis.

Limitations remain. The cohort was hospital-based and highly selected: only pathologically confirmed BI-RADS 4–5 lesions larger than 5 millimeters were included, excluding prostheses, poor-quality videos, and other BI-RADS categories. No asymptomatic screening population was tested, and screening settings — with lower disease prevalence and smaller incidental lesions — may behave very differently. The authors themselves urge prospective validation in screening cohorts before broader deployment. Still, the study marks a conceptual shift in medical AI: from judging isolated snapshots to understanding the full temporal narrative of a scan, the way a human expert does. If subsequent trials confirm these findings, the humble ultrasound video — long a passive record of a scan — may become an active diagnostic instrument, and the gap between novice and veteran sonographers may finally begin to close.

Subject of Research: Deep learning analysis of ultrasound videos for classifying benign and malignant BI-RADS 4–5 breast lesions

Article Title: AI-enhanced ultrasound video analysis for BI-RADS 4–5 breast lesions: a multicenter study of diagnostic performance

Article References: Li, W.-B., Du, Z.-C., Huang, W.-H., Qiu, R.-Y., Chen, Y.-C., Lu, L.-S., Ji, Z.-Y., Guo, L.-H., He, M.-Q., Wang, L.-S., & Zhang, G.-J. (2026). AI-enhanced ultrasound video analysis for BI-RADS 4–5 breast lesions: a multicenter study of diagnostic performance. Holistic Integrative Oncology, 5(1), Article 80. https://doi.org/10.1007/s44178-026-00302-8

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00302-8

Keywords: artificial intelligence, deep learning, breast cancer, ultrasound, BI-RADS, computer-aided diagnosis, multi-instance learning, medical imaging, diagnostic accuracy, oncology, radiology, multicenter study

The patient’s story began in September 2019, when she first noticed a small, mobile mass in her left breast measuring roughly 2 by 1 by 1 centimeters. The lump caused no tenderness, no skin changes, and no nipple discharge, so she initially ignored it. Two months later, at another hospital, mammography and breast color Doppler ultrasound both classified the lesion as Breast Imaging Reporting and Data System category III—a designation implying a probably benign finding with less than a 2 percent risk of malignancy—and recommended routine follow-up. Over the following three months, however, the mass grew dramatically, reaching approximately 4 by 3 by 2 centimeters and occasionally becoming painful. That rapid enlargement prompted her admission to the reporting hospital in March 2020, where physical examination found a smooth, moderately firm, mobile mass without skin ulceration, nipple retraction, or enlarged lymph nodes in the armpits.

The imaging workup revealed a striking contradiction. Mammography at the outside hospital had graded the lesion BI-RADS 3, but repeat ultrasound at the new institution upgraded it to BI-RADS 4A, indicating a small but real suspicion of malignancy of roughly 2 to 10 percent. The ultrasound showed a solid, irregular, lobulated mass at the 11 o’clock position of the left breast, containing strip-like blood flow signals with low-resistance arterial waveforms—a vascular pattern that would later prove telling. Chest computed tomography confirmed a mound-shaped mass in the upper inner quadrant with an unclear boundary on plain scan. Meanwhile, laboratory results were entirely unremarkable: blood counts, liver and kidney function, and coagulation were normal, and the tumor markers CEA, CA 15-3, CA 125, and CA 19-9 all sat comfortably within reference ranges. The team noted that breast angiosarcomas frequently produce negative results on routine malignant tumor markers, rendering these tests of extremely low diagnostic value in such cases.

Reconciling the discordant imaging grades against current standards proved decisive in shaping the diagnostic pathway. According to the fifth edition of the American College of Radiology BI-RADS criteria, the 2024 National Comprehensive Cancer Network breast guidelines, and the 2021 Chinese Society of Breast Surgery fibroadenoma guideline, the combined clinical and imaging picture did not meet formal indications for contrast-enhanced breast MRI, bone scintigraphy, or percutaneous core needle biopsy. Those tools are reserved for high-risk patients, radiologically indeterminate lesions with definite malignant suspicion, or preoperative staging of confirmed cancers—not for discordant low-grade BI-RADS 4A lesions accompanied by benign mammographic findings. The clinicians therefore concluded the lesion was highly suggestive of a benign fibroadenoma and recommended elective surgical resection, a decision that conformed to guidelines but, as they later acknowledged, failed to account specifically for the lesion’s rapid growth.

Surgery took place on March 11, 2020, under general anesthesia, and it was here that the case pivoted. Intraoperative frozen section pathology suggested a spindle cell tumor tending toward a vascular origin, with definitive classification awaiting immunohistochemistry. Because the team had rarely encountered such lesions, and because committing to a radical operation on an uncertain diagnosis carried obvious risks, they consulted the patient’s family and made the unusual decision to temporarily suspend the surgery. Final pathology returned on March 17: the excised nodule, about 3.5 by 2.5 by 1.8 centimeters, was packed with anastomosing slit-like vascular channels infiltrating breast lobules and fat, lined by a single layer of endothelial cells showing mild nuclear atypia and occasional mitoses. Immunohistochemistry showed strong positivity for the endothelial markers CD34, CD31, and FLT-1, along with vimentin and smooth muscle actin, while epithelial markers EMA and broad-spectrum cytokeratins were negative. The Ki-67 proliferation index was 25 percent. The diagnosis was low-grade angiosarcoma.

To be certain, the slides were sent for outside consultation at a higher-level hospital, which returned on March 20 with a confirmatory and slightly more concerning picture: part of the tumor showed the mild, well-formed vascular pattern, but other areas were solid, densely cellular, and pleomorphic with mitotic figures, extensive hemorrhage, and necrosis. The consultants recommended optional molecular testing for c-Myc, KDR, and VEGFR2—markers mainly useful for distinguishing primary from radiation-associated secondary angiosarcoma—though the team ultimately judged these unnecessary given the clear histopathological diagnosis. After a hospital-wide multidisciplinary team discussion, the patient underwent completion resection of the remaining left breast tissue plus sentinel lymph node biopsy on March 23. Frozen section confirmed an R0 resection with clear microscopic margins and no sentinel node metastasis, and final pathology found no residual cancer in the surrounding breast, nipple, skin, or basal margin, with zero of three sentinel nodes involved.

The staging and treatment decisions reflected the peculiar biology of this tumor. Under the eighth edition of the AJCC staging system, the disease was classified as stage I (T1N0M0). Unlike conventional breast carcinoma, which spreads chiefly through lymphatic channels, primary breast angiosarcoma metastasizes predominantly through the bloodstream, seeding the liver, lungs, bone, and brain early, while regional lymph node involvement is extremely rare. Routine axillary lymph node dissection is therefore not recommended, and the team performed only a sentinel node biopsy for accurate staging, sparing the patient unnecessary surgical trauma. Beginning April 3, 2020, she received eight cycles of adjuvant chemotherapy—four of doxorubicin plus cyclophosphamide followed by four of paclitaxel—completed without significant adverse reactions by the end of July 2020.

The prognosis data assembled in the report underscore why vigilance matters. Secondary angiosarcoma of the breast, which typically arises after radiotherapy for breast cancer or in the setting of chronic lymphedema known as Stewart-Treves syndrome, carries a five-year overall survival of roughly 38 percent with a recurrence-free survival of about 43 months. Primary breast angiosarcoma fares somewhat better, with a five-year overall survival near 49 percent and recurrence-free survival of about 44 months, with outcomes tied to stage, histological grade, and tumor size. Diagnosis remains the hardest part: mammography yields false negatives in roughly a third of patients because of dense glandular tissue, ultrasound lacks specificity despite showing heterogeneous masses and increased blood flow, and MRI—considered the most useful non-invasive modality—reveals heterogeneous signals and avid contrast enhancement reflecting the tumor’s rich vascularity. Core needle biopsy, the usual route to preoperative diagnosis, frequently returns false negatives because of bleeding risk and marked tumor heterogeneity, meaning most cases are only confirmed after surgery.

Emerging evidence is beginning to reshape treatment thinking for this orphan disease. A meta-analysis by Skarentzos and colleagues found that adjuvant chemotherapy significantly improved survival in primary breast angiosarcoma, with a hazard ratio of 0.11, while radiotherapy showed no survival benefit. Anthracycline-, ifosfamide-, and paclitaxel-containing regimens are considered the most effective, though the disease’s rarity means evidence rests on small series and case reports rather than large trials. Genomic research is opening new doors: the Angiosarcoma Project, a patient-partnered genomic study, has suggested that inhibitors of the phosphatidylinositol-3-kinase catalytic alpha subunit could become targeted therapeutic options, and next-generation sequencing may eventually identify actionable alterations in tumors like this one.

Follow-up tells the story’s encouraging end. The patient was re-examined at one, three, and six months after surgery, then roughly every six months, with breast ultrasound, chest CT, abdominal and urinary tract ultrasound, blood counts, and tumor markers. As of November 5, 2025—more than five years after her diagnosis—she showed no sign of local recurrence or distant metastasis. The authors argue their case carries three broader lessons: that a rapidly growing, benign-appearing breast mass in a postmenopausal woman deserves heightened suspicion even when guidelines do not mandate advanced imaging; that suspending surgery when frozen section reveals something unexpected can preserve diagnostic accuracy and prevent undertreatment; and that multidisciplinary collaboration, staged surgery, and guideline-conformant individualized therapy can deliver excellent long-term outcomes even in a cancer this rare and this easy to miss.

Subject of Research: Primary breast angiosarcoma diagnosis and treatment in a postmenopausal woman

Article Title: Primary Mammary Angiosarcoma in a Postmenopausal Woman: A Case Report

Article References: Liu, T., Xu, Y., Wang, Q., Yu, S., Lin, H., Ye, Z., Zheng, Z., Yang, C., & Hong, T. (2026). Primary Mammary Angiosarcoma in a Postmenopausal Woman: A Case Report. Cancer Reports, 9(9), Article e70653. https://doi.org/10.1002/cnr2.70653

Image Credits: AI Generated

DOI: 10.1002/cnr2.70653

Keywords: breast angiosarcoma, primary breast angiosarcoma, postmenopausal, BI-RADS, fibroadenoma, frozen section pathology, immunohistochemistry, CD31, sentinel lymph node biopsy, adjuvant chemotherapy, R0 resection, rare cancer