Torres, F.F.TorresEscalante-Ramirez, B.B.Escalante-RamirezOlveres, J.J.OlveresPING-LANG YEN2021-02-202021-02-202019https://www.scopus.com/inward/record.url?eid=2-s2.0-85076107075&partnerID=40&md5=50061baf6b357978a7b02bf55a338bc1https://scholars.lib.ntu.edu.tw/handle/123456789/549067Breast ultrasound has become one of the most important and effective modalities for early detection of breast cancer and it is most suitable for large scale breast cancer screening and diagnosis in low-resource countries. Breast lesion detection is a crucial step in the development of Computer Aided Diagnosis and Surgery systems based on ultrasound images, since it can be used as a seed point to subsequently initialize segmentation methods such as region growing, snakes or level-sets. Because of inherent artifacts of the ultrasound images, such as speckle, acoustic shadows and blurry edges, the detection of lesions is not an easy task. In this work we propose a machine learning based approach to locate lesions in breast ultrasound images. This approach consists on the classification of image pixels as lesion or background with a Random Forest optimized with genetic algorithms to generate candidate regions. After pixel classification the method chooses the correct lesion region by discriminating false positives using a new proposed probability approach. The pixel classification and region discrimination steps are compared with other methods, showing better results in the detection of lesions. The lesion detection was evaluated using the True Positive Fraction and the False Positives per image, having results of 84.4% and 15.6% respectively. ? 2019, Springer Nature Switzerland AG.Breast lesion; Genetic algorithms; Random Forest; Ultrasound[SDGs]SDG3[SDGs]SDG5Computer aided diagnosis; Decision trees; Diseases; Genetic algorithms; Image analysis; Image segmentation; Learning algorithms; Machine learning; Numerical methods; Pattern recognition; Pixels; Ultrasonic applications; Ultrasonics; Breast cancer screening; Breast lesion; Breast lesion detection; Breast ultrasound images; Early detection of breast cancer; Machine learning approaches; Random forests; True positive fractions; Medical imagingconference paper10.1007/978-3-030-31332-6_262-s2.0-85076107075