Showing 1–1 of 1 results for author: Jinnai, Y

Blind Signal Separation for Fast Ultrasound Computed Tomography

Authors: Takumi Noda, Yuu Jinnai, Naoki Tomii, Takashi Azuma

Abstract: Breast cancer is the most prevalent cancer with a high mortality rate in women over the age of 40. Many studies have shown that the detection of cancer at earlier stages significantly reduces patients' mortality and morbidity rages. Ultrasound computer tomography (USCT) is considered as a promising screening tool for diagnosing early-stage breast cancer as it is cost-effective and produces 3D imag… ▽ More Breast cancer is the most prevalent cancer with a high mortality rate in women over the age of 40. Many studies have shown that the detection of cancer at earlier stages significantly reduces patients' mortality and morbidity rages. Ultrasound computer tomography (USCT) is considered as a promising screening tool for diagnosing early-stage breast cancer as it is cost-effective and produces 3D images without radiation exposure. However, USCT is not a popular choice mainly due to its prolonged imaging time. USCT is time-consuming because it needs to transmit a number of ultrasound waves and record them one by one to acquire a high-quality image. We propose FastUSCT, a method to acquire a high-quality image faster than traditional methods for USCT. FastUSCT consists of three steps. First, it transmits multiple ultrasound waves at the same time to reduce the imaging time. Second, it separates the overlapping waves recorded by the receiving elements into each wave with UNet. Finally, it reconstructs an ultrasound image with a synthetic aperture method using the separated waves. We evaluated FastUSCT on simulation on breast digital phantoms. We trained the UNet on simulation using natural images and transferred the model for the breast digital phantoms. The empirical result shows that FastUSCT significantly improves the quality of the image under the same imaging time to the conventional USCT method, especially when the imaging time is limited. △ Less

Submitted 27 April, 2023; originally announced April 2023.