2020/09/23 by Gavriel Habib, Nahum Kiryati, Habib, Gavriel +15
Computer Science · Medicine · #AI in cancer detection #COVID-19 diagnosis using AI #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Radiomics and Machine Learning in Medical Imaging #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2009.11009
openalex publication_date 2020/09/23 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Mammography and ultrasound are extensively used by radiologists as\ncomplementary modalities to achieve better performance in breast cancer\ndiagnosis. However, existing computer-aided diagnosis (CAD) systems for the\nbreast are generally based on a single modality. In this work, we propose a\ndeep-learning based method for classifying breast cancer lesions from their\nrespective mammography and ultrasound images. We present various approaches and\nshow a consistent improvement in performance when utilizing both modalities.\nThe proposed approach is based on a GoogleNet architecture, fine-tuned for our\ndata in two training steps. First, a distinct neural network is trained\nseparately for each modality, generating high-level features. Then, the\naggregated features originating from each modality are used to train a\nmultimodal network to provide the final classification. In quantitative\nexperiments, the proposed approach achieves an AUC of 0.94, outperforming\nstate-of-the-art models trained over a single modality. Moreover, it performs\nsimilarly to an average radiologist, surpassing two out of four radiologists\nparticipating in a reader study. The promising results suggest that the\nproposed method may become a valuable decision support tool for breast\nradiologists.\n