Tomosynthesis-only Detected Architectural Distortion: A Meta-analysis of 1361 Lesions

To quantify malignancy, high-risk, and benign proportions in architectural-distortion (AD) lesions that are visible only on digital breast tomosynthesis (DBT) and to assess between-study heterogeneity and publication bias.

A systematic search from January 2010 to March 2025 yielded 14 single-arm cohorts (1361 lesions). Two reviewers extracted per-study counts of malignant, high-risk, and benign pathology. Pooled proportions were calculated with DerSimonian-Laird random-effects models; heterogeneity was expressed as I² and τ². Funnel plots and Egger regression tested for small-study effects. Sensitivity analyses excluded cohorts with < 50 lesions.

For tomosynthesis only detected AD, malignancy accounted for 37% (95% CI 26-48; I² ≈ 94) and benign outcomes for 41% (95% CI 28-54; I² ≈ 96). By complement, non‑benign pathology (malignant + high‑risk) was 59% (95% CI 46-73) CI 28 - 54%; I² = 96%) of the lesions. Pooling malignant and high-risk categories yielded a "non-benign" rate of 59% (95% CI 46 - 74%). Egger tests were non-significant for benign and high-risk outcomes (p = 0.879 and 0.770, respectively) and borderline for malignancy (p = 0.050). Restricting the analysis to the cohorts with ≥ 50 lesions changed the pooled malignancy rate only modestly (39%) and reduced the high-risk lesions to 4%; heterogeneity remained extreme (I² > 90%). Variability was linked to biopsy gauge (vacuum-assisted vs 14-G core), ultrasound-correlate policy, and referral setting. Funnel plots were largely symmetric; no outcome showed strong evidence of study bias.

Approximately one in three DBT-only detected AD lesions is malignant, which supports the recommendation of a biopsy rather than conservative management with a short-interval follow-up.