Managing breast complaints during pregnancy or lactation presents unique clinical difficulties for Indian clinicians. Fortunately, the development of breast ultrasound AI provides a promising way to support decision-making in these scenarios. A recent retrospective study evaluated the accuracy of an AI-based tool in assessing breast lesions. Consequently, this technology could help clinicians distinguish benign findings from rare malignancies.
Evaluating Breast Ultrasound AI Performance
Researchers at two academic medical centers studied 504 patients with 639 breast ultrasound findings from 2018 to 2021. Specifically, they compared the AI tool’s BI-RADS classifications against human radiologist interpretations. Surprisingly, both the radiologists and the AI tool achieved an identical sensitivity of 80% for malignancy. Specifically, they both correctly identified four of five malignant findings. However, the AI tool recommended significantly more biopsies for benign lesions than the human radiologists. Therefore, the AI tool showed a lower specificity in this cohort.
Clinical Implications for Practice
In clinical practice, the lower specificity of the AI tool presents challenges. In particular, utilizing this tool blindly could trigger unnecessary biopsies for benign lactational changes. Nevertheless, AI can still act as a highly valuable secondary check to ensure clinicians do not miss rare cancers. Furthermore, Indian healthcare settings can benefit from AI decision support where expert breast radiologists are scarce. Consequently, clinicians must always combine AI assessments with their clinical judgment to avoid over-investigating benign masses.
Frequently Asked Questions
Q1: What was the diagnostic sensitivity of the breast ultrasound AI tool?
Both the AI tool and human radiologists achieved an identical sensitivity of 80%. Specifically, they correctly identified four of five malignant findings.
Q2: Did the AI tool recommend more biopsies than human radiologists?
Indeed, the AI tool recommended significantly more biopsies for benign lesions. Consequently, this resulted in a lower specificity compared to human radiologists.
Q3: Should clinicians rely solely on AI tools during pregnancy and lactation?
In fact, benign lactational changes can easily mimic abnormal findings. Consequently, clinicians must always combine AI findings with expert human interpretation to avoid unnecessary biopsies.
References
- Dwan D et al. Diagnostic accuracy of an artificial intelligence-based breast ultrasound tool in pregnant and lactating patients. Eur Radiol. 2026 Jun 10. doi: 10.1007/s00330-026-12659-5. PMID: 42268305.
- Sabate JM et al. Radiologic Evaluation of Breast Disorders Related to Pregnancy and Lactation. Radiographics. 2007;27(6):1705-1724. doi: 10.1148/rg.276065153.