Segmentation and three-dimensional reconstruction of lesions using the automated breast volume scanner (ABVS)

Background: Ultrasound is an effective tool for breast cancer diagnosis. However, its relatively low image quality makes small lesion analysis challenging. This promotes the development of tools to help clinicians in the diagnosis. Methods: We propose a method for segmentation and three-dimensional (3D) reconstruction of lesions from ultrasound images acquired using the automated breast volume scanner (ABVS). Segmentation and reconstruction algorithms are applied to obtain the lesion's 3D geometry. A total of 140 artificial lesions with different sizes and shapes are reconstructed in gelatin-based phantoms and biological tissue. Dice similarity coefficient (DSC) is used to evaluate the reconstructed shapes. The algorithm is tested using a human breast phantom and clinical data from six patients. Results: DSC values are 0.86 ± 0.06 and 0.86 ± 0.05 for gelatin-based phantoms and biological tissue, respectively. The results are validated by a specialized clinician. Conclusions: Evaluation metrics show that the algorithm accurately segments and reconstructs various lesions.