Various deep learning methods have been proposed to segment breast lesions from ultrasound images. However, similar intensity distributions, variable tumor morphologies and blurred boundaries present challenges for breast lesions segmentation, especially for malignant tumors with irregular shapes. Considering the complexity of ultrasound images, we develop an adaptive attention U-net (AAU-net) to segment breast lesions automatically and stably from ultrasound images. Specifically, we introduce a hybrid adaptive attention module (HAAM), which mainly consists of a channel self-attention block and a spatial self-attention block, to replace the traditional convolution operation. Compared with the conventional convolution operation, the design of the hybrid adaptive attention module can help us capture more features under different receptive fields. Different from existing attention mechanisms, the HAAM module can guide the network to adaptively select more robust representation in channel and space dimensions to cope with more complex breast lesions segmentation. Extensive experiments with several state-of-the-art deep learning segmentation methods on three public breast ultrasound datasets show that our method has better performance on breast lesions segmentation. Furthermore, robustness analysis and external experiments demonstrate that our proposed AAU-net has better generalization performance in the breast lesion segmentation. Moreover, the HAAM module can be flexibly applied to existing network frameworks. The source code is available on https://github.com/CGPxy/AAU-net.

Milling is one of the most popular forms of hard tissue removal process, it is sometimes high-risk because there may be very small distance between the milling trajectory and vital anatomy. This paper introduces a vibration signal acquiring and processing method to identify different types of milling states in robot-assisted orthopedics. During milling process, the tissue vibration signal measured by a laser displacement sensor, and the acceleration signal of the operation power device recorded by an accelerometer, are decomposed by the lifting wavelet packet transform, so the harmonic component whose frequency is an integer times of the spindle frequency of the operation power device is extracted. In consideration of that the tissue vibration amplitude varies with the milling force, the relative magnitude of the force is estimated from the acceleration signal, and then the wavelet energy of the tissue vibration signal is divided by that of the acceleration signal to reduce the influence of the force disturbance. The compensated wavelet energy is subsequently used as input vector to a support vector machine for discriminating different milling states. Experimental results on porcine spines demonstrate that the proposed method can successfully discriminate the tissues in the operation field: the success rate achieves 100% for the vertebra being milled and the spinal cord, for the adjacent bony structure and the muscle the success rate is about 90%.

In order to enhance the safety of the milling operation, we develop a condition monitoring method by means of analyzing the bone vibration signal in robot-assisted spine surgery. Since the behavior of the bone being cut changes with the tissue removal, an analytical method is proposed for modeling of varying bone dynamics, and then, it is proved that the vibration amplitude of the bone indicates its status change. During the real milling process, the vibration signal of the bone is recorded by a noncontact laser displacement sensor, and we only pay attention to the harmonic component whose frequency is an integer times of the spindle frequency of the milling device. The wavelet packet transform and the adaptive linear element are performed to estimate the harmonic amplitude from the signal to correlate milling condition. Considering that the bone vibration also varies with the milling force, the relative magnitude of the force is measured from the wavelet coefficients in low-frequency subband, and subsequently, the estimated harmonic amplitude is compensated to reduce the influence of the force. The robot milling experiments in sheep spines are carried on to verify the effectiveness of the condition monitoring procedure. The experimental results are in substantial accordance with the conclusions from the dynamic model, and thus, the robot can improve the safety of the operation with the proposed method.

Physiological studies in humans proved that the central nervous system automatically integrates the information from different sensory systems. Inspired by the integration mechanism, this study aims to yield accurate perception of the surgical field during milling operation. The dynamic model of surgical milling process is developed theoretically, and the model considers the influence of the unbalance in the rotating machine and the time-variant cutting force. On the basic of the theoretical analysis, the monitoring of the milling process can be realized to determine the state of the operation. An accelerometer and a microphone are used to measure the acceleration of the milling device and the sound pressure generated during bone milling separately, and then the correlation coefficient between the harmonics in the two signals is fed into the artificial neural network. The output of the artificial neural network is used to determine whether the milling device is cutting the cortical bone, the cancellous bone, the muscle or nothing (i.e. idle running). The experimental results in in vitro porcine spines show that the proposed method is very sensitive to the phenomenon of tissue wrapping in the surgical field, and it can identify the predetermined states correctly.