# Suspicion for Sarcoma: Clinical Presentation, Multi-Modality Imaging Evaluation, and Ultrasound Artificial Intelligence-Based Decision Support

## TL;DR

This study examines how breast sarcomas appear on different imaging techniques and shows that AI can help detect them with high accuracy.

## Contribution

The study introduces the use of ultrasound AI decision support for detecting breast sarcomas and characterizes their imaging features.

## Key findings

- Breast sarcomas commonly appear as irregular-shaped masses with non-circumscribed margins on imaging.
- Ultrasound AI decision support accurately identified 93.8% of breast sarcoma lesions as suspicious.
- MRI showed heterogeneously enhancing masses or isolated skin enhancement in most cases.

## Abstract

Breast sarcomas are rare and aggressive. Our study aims to better characterize the clinical presentation, histology, and imaging features of breast sarcomas on mammography, ultrasound, and MRI, in addition to analyzing the effectiveness of ultrasound AI decision support (DS) in detecting breast sarcomas. A retrospective review from 2008–2024 yielded 18 patients with histologically proven breast sarcomas with imaging available for review. Mammography was available for 13 lesions, ultrasound for 19 lesions, and MRI for 9 lesions. The most common presentation of breast sarcoma was as an irregular-shaped mass with non-circumscribed margins on mammography, ultrasound, and MRI, the latter with heterogenous enhancement. Ultrasound AI DS accurately identified 15 out of 16 (93.8%) breast sarcoma lesions seen on ultrasound as suspicious. Awareness of how breast sarcomas can present across imaging modalities while using AI DS as an aid may help radiologists in making the correct diagnosis of this rare and aggressive disease.

Background/Objective: This study aims to better characterize the clinical presentation, histology, and imaging features of breast sarcomas on mammography, ultrasound, and MRI, in addition to analyzing the effectiveness of AI DS in detecting breast sarcomas. Methods: A retrospective review from 2008–2024 yielded 18 patients with histologically proven breast sarcomas with imaging available for review. Mammography was available for 13 lesions, ultrasound for 19 lesions, and MRI for 9 lesions. Imaging features were classified according to the BI-RADS 5th edition lexicon. Images were reviewed by two radiologists, and consensus was obtained regarding imaging features. AI DS was retrospectively applied to the breast masses identified on ultrasound. Data analysis was performed using descriptive statistics. Results: 17 females and 1 male were included in this study. Mammographic findings varied from solitary masses (3/13 [23.1%]), asymmetries (3/13 [23.1%]), architectural distortion (1/13 [7.7%]), skin thickening (3/13 [23.1%]), focal asymmetry with calcifications (1/13 [7.7%]), or no suspicious findings (2/13 [15.4%]). Sonography often revealed masses with an irregular shape (13/16 [81.2%]), non-circumscribed margins (15/16 [93.7%]), hypoechoic echo pattern (10/16 [62.5%]), and vascular flow (12/16 [75%]). MRI showed heterogeneously enhancing masses (6/9 [66.7%]) or isolated skin enhancement (3/9 [33.3%]). AI DS analyzed 16 masses on ultrasound and identified 15 (93.8%) as suspicious. Conclusions: Breast sarcomas had a variable appearance on breast imaging, ranging from a solitary mass to isolated skin findings. Awareness of how breast sarcomas can present across imaging modalities while using AI DS as an aid may help radiologists in making the correct diagnosis of this rare and aggressive disease.