新生児集中治療室(neonatal intensive care unit,NICU)の対象になる低出生体重児は,重篤な状態に陥りやすい状態にあり,瞬時心拍数・動脈血酸素飽和度・呼吸数などの生体信号のモニタリングが常時行われている。しかし,モニタリングされている生体信号は瞬時的な値しかほとんど注目されておらず,長期的にみたときにどのような変動がみられるのかはあまり知られていない。本研究ではNICUに入院中の低出生体重児95名について生体信号を長期的に記録保存し、各データの変動や出生時体重・在胎期間などのパラメータとの相関について検討した。その結果、退院時に71.6%の被験者が在胎期間から約244日、SPO2が96.2%以上、平均呼吸数が52(回/分)以下の範囲の値をとることがわかった。
In the diagnostic ultrasound scan images, granular structures appear in the images. It is necessary to develop more robust despeckling techniques for enhance the ultrasound medical images. A new filter that balances between speckle reduction and edge (detail) preservation is proposed. The new approach adjusts the filter’s window size and varying the weights used in median filter by utilizing the local statistics inside the filter.
The speckle noise has a different characteristic comparison with natural images and its evaluation can be defined by an investigator who observes the normal part and abnormal on sonography image. Because of the speckle noise is one of the major sources of image quality degradation and its special scanning conversion processing, a lot of conventional noise reduction method is not enough for speckle noise reduction. In this paper a new method to enhance ultrasound images is proposed. The new approach attempts to confirm the adaptive threshold value of US image by an artificial speckled US image modeling. The multi-layer wavelet filtering method is proposed by wavelet decomposition based adaptive thresholding.
A diagnosis for diseases by using an ultrasound images is difficult to determine that the tissue is good or not good. For this reason, the reduction of speckle is important to check the disease. In this paper, the purpose is to reduce speckle noise in ultrasound images. We preprocessed the images by using speckle reducing anisotropic diffusion (SRAD) which is the edge-sensitive diffusion for speckled images. It is important to decide the homogeneous area of speckle scale function to get the better result.
We proposed preprocessing method based on the cellular neural network to reduce noise. We present an active contour noise decrease and a clear boundary through proposed preprocessing method. We find out tumor boundary and shape extraction using a level set based on active contour. The proposed preprocessing method has been applied to real medical ultrasound images with promising detected active contour results.
Segmentations of ultrasound images are the basis of automatic detection and characterization of breast cancer computer-aided diagnosis systems. Many segmentation methods have been developed and a challenge of this work is that we have to accommodate artifacts, such as the speckle, as well as the fuzziness of lesion boundaries. In this paper, we present an automatic texture-based segmentation algorithm for breast ultrasound images. The framework of this method can be depicted by two parts: 1) extracting texture energy features, 2) doing fuzzy c-mean clustering in texture domain.