Each breast cancer has its unique spatial shape, but the clinical importance and the underlying mechanism for the three-dimensional tumor shapes are mostly unknown. We collected the data on the three-dimensional tumor size and tumor volume data of invasive breast cancers from 2,250 patients who underwent surgery between Jan 2000 and Jul 2007. The degree of tumor eccentricity was estimated by using the difference between the spheroid tumor volume and ellipsoid tumor volume (spheroid-ellipsoid discrepancy, SED). In 41 patients, transcriptome and exome sequencing data obtained. Estimation of more accurate tumor burden by calculating ellipsoid tumor volumes did not improve the outcome prediction when compared to the traditional longest diameter measurement. However, the spatial tumor eccentricity, which was measured by SED, showed significant variation between the molecular subtypes of breast cancer. Additionally, the degree of tumor eccentricity was associated with well-known prognostic factors of breast cancer such as tumor size and lymph node metastasis. Transcriptome data from 41 patients showed significant association between MMP13 and spatial tumor shapes. Network analysis and analysis of TCGA gene expression data suggest that MMP13 is regulated by ERBB2 and S100A7A. The present study validates the usefulness of the current tumor size method in determining tumor stages. Furthermore, we show that the tumors with high eccentricity are more likely to have aggressive tumor characteristics. Genes involved in the extracellular matrix remodeling can be candidate regulators of the spatial tumor shapes in breast cancer.