Related Experiment Video
Updated: Nov 24, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
A risk stratification algorithm for lesions of uncertain malignant potential diagnosed by vacuum-assisted breast
Paola Clauser1, Panagiotis Kapetas1, Alexander Stöttinger1
1Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna, Waehringer Guertel 18-20, Vienna, Austria.
Purpose:
To investigate a risk stratification strategy for lesions of uncertain malignant potential (B3) diagnosed by vacuum-assisted breast biopsy (VABB) of mammographic microcalcifications.
Methods:
Patients who underwent VABB for microcalcification-only lesions with a diagnosis of B3 and subsequent surgery were included in this retrospective, IRB-approved study. Seventy-six B3-lesions (final histology: 66 benign, 10 malignant) were included (Tr). Data on B3 lesion type and presence of atypia, microcalcification characteristics (BI-RADS), removal at biopsy and concomitant lesions were collected. After univariate analysis (Chi-square test), data were combined into a risk stratification algorithm by using a ten-fold, cross-validated Classification and Regression Tree analysis (CRT). The algorithm was tested on a testing dataset (Te) of 23 B3-lesions (six malignant, 17 benign).
Results:
Malignancy was more frequent in women with a concomitant cancer (P < 0.001) and highly suspicious microcalcifications (P < 0.001). The CRT algorithm retained three characteristics: morphology; presence of atypia; presence of concomitant cancer. The algorithm identified 25/76 (32.9 %,Tr) and 6/23 (26.1 %,Te) lesions at low risk of malignancy. No malignant cases were identified at surgery (0/31). There were 3/76 (3.9 %,Tr) and 1/23 (4.3 %,Te) lesions assigned as high-risk by the algorithm and confirmed at surgery (4/4). In the remaining lesions (48/76, 63.1 %,Tr; 16/23, 69.6 %,Te), malignancy rates varied between 9% and 88.4 %; thus, surgery could not have been avoided.
Conclusion:
We constructed and tested a risk stratification algorithm for B3 microcalcifications, including clinical, imaging, and pathological features, to assign probabilities of malignancy, which has the potential to reduce unnecessary surgeries.
