Syahril Siregar, Zahra Azizah, Siti Julia
—This study develops an interpretable diagnostic model based on shape-derived features to distinguish benign from malignant breast lesions in ultrasound imaging. A dataset of 437 benign and 210 malignant cases with expert-annotated lesion masks was analyzed. Six descriptors—axis ratio, circularity, solidity, extent, elongation, and eccentricity—were extracted and classified using Partial Least Squares Discriminant Analysis (PLS-DA). Model performance was evaluated with 5-fold cross-validation. The PLS-DA achieved 92.4% accuracy and an F1-score of 85.7%. Score plots revealed clear separation, with benign lesions showing higher circularity, solidity, extent, elongation, and eccentricity, reflecting regular, well-defined shapes. Elongation and solidity emerged as the most discriminative features, supported by Variable Importance in Projection (VIP) scores. The results demonstrate that shape-based PLS-DA provides both predictive power and interpretability, offering a transparent alternative to black-box AI. This approach may reduce invasive procedures and enhance diagnostic confidence. Future work should incorporate texture or radiomic features to improve generalizability. ©2025 IEEE.
Department of Physics, FMIPA, Universitas Indonesia Depok, Indonesia; Department of Computer and Informatics Engineering, Politeknik Negeri Jakarta, Depok, Indonesia; Department of Physics, FMIPA, State University of Jakarta, Jakarta, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock