In this paper, a novel approach of high-accuracy calibration (HAC) method is employed to improve the resolution of tumor detection within a fibro-glandular breast model, and also an improved 3D back-projection approach to scan each focal point inside of the breast is presented. For these purposes, a simulated hemispherical setup of a multi-static array with a modified UWB bowtie antenna is applied around the breast. The superiority of the proposed HAC method is that all-time delays of multi-static channel paths are taken into account at preprocessing of reflected signals, and therefore the time location of tumor response can be estimated accurately in the late stage of recorded signals. As a result, stronger signals are obtained to detect the location of the tumor with higher spatial accuracy. By using the improved 3D back-projection method, a better approximation of transmission channel paths based on Fermat's principle is achieved. A realistic breast model is proposed with two cases of single and twin spherical tumors. Then, to validate the efficiency of the proposed HAC method to detect the time-dependent tumor location, several scenarios are studied in the mentioned model. Quantitative metrics of successfully reconstructed tumor (with a small radius of 7 mm) images prove the ability of the proposed imaging method for early-stage breast cancer detection.