A two-dimensional video based quantification method and clinical application research of motion disorders / 生物医学工程学杂志

The increasing prevalence of the aging population, and inadequate and uneven distribution of medical resources, have led to a growing demand for telemedicine services. Gait disturbance is a primary symptom of neurological disorders such as Parkinson's disease (PD). This study proposed a novel approach for the quantitative assessment and analysis of gait disturbance from two-dimensional (2D) videos captured using smartphones. The approach used a convolutional pose machine to extract human body joints and a gait phase segmentation algorithm based on node motion characteristics to identify the gait phase. Moreover, it extracted features of the upper and lower limbs. A height ratio-based spatial feature extraction method was proposed that effectively captures spatial information. The proposed method underwent validation via error analysis, correction compensation, and accuracy verification using the motion capture system. Specifically, the proposed method achieved an extracted step length error of less than 3 cm. The proposed method underwent clinical validation, recruiting 64 patients with Parkinson's disease and 46 healthy controls of the same age group. Various gait indicators were statistically analyzed using three classic classification methods, with the random forest method achieving a classification accuracy of 91%. This method provides an objective, convenient, and intelligent solution for telemedicine focused on movement disorders in neurological diseases.

A generative adversarial network-based unsupervised domain adaptation method for magnetic resonance image segmentation / 生物医学工程学杂志

Intelligent medical image segmentation methods have been rapidly developed and applied, while a significant challenge is domain shift. That is, the segmentation performance degrades due to distribution differences between the source domain and the target domain. This paper proposed an unsupervised end-to-end domain adaptation medical image segmentation method based on the generative adversarial network (GAN). A network training and adjustment model was designed, including segmentation and discriminant networks. In the segmentation network, the residual module was used as the basic module to increase feature reusability and reduce model optimization difficulty. Further, it learned cross-domain features at the image feature level with the help of the discriminant network and a combination of segmentation loss with adversarial loss. The discriminant network took the convolutional neural network and used the labels from the source domain, to distinguish whether the segmentation result of the generated network is from the source domain or the target domain. The whole training process was unsupervised. The proposed method was tested with experiments on a public dataset of knee magnetic resonance (MR) images and the clinical dataset from our cooperative hospital. With our method, the mean Dice similarity coefficient (DSC) of segmentation results increased by 2.52% and 6.10% to the classical feature level and image level domain adaptive method. The proposed method effectively improves the domain adaptive ability of the segmentation method, significantly improves the segmentation accuracy of the tibia and femur, and can better solve the domain transfer problem in MR image segmentation.

Robust tracking of dexterous continuum robots: Fusing FBG shape sensing and stereo vision.

Robust and efficient tracking of continuum robots is important for improving patient safety during space-confined minimally invasive surgery, however, it has been a particularly challenging task for researchers. In this paper, we present a novel tracking scheme by fusing fiber Bragg grating (FBG) shape sensing and stereo vision to estimate the position of continuum robots. Previous visual tracking easily suffers from the lack of robustness and leads to failure, while the FBG shape sensor can only reconstruct the local shape with integral cumulative error. The proposed fusion is anticipated to compensate for their shortcomings and improve the tracking accuracy. To verify its effectiveness, the robots' centerline is recognized by morphology operation and reconstructed by stereo matching algorithm. The shape obtained by FBG sensor is transformed into distal tip position with respect to the camera coordinate system through previously calibrated registration matrices. An experimental platform was set up and repeated tracking experiments were carried out. The accuracy estimated by averaging the absolute positioning errors between shape sensing and stereo vision is 0.67±0.65 mm, 0.41±0.25 mm, 0.72±0.43 mm for x, y and z, respectively. Results indicate that the proposed fusion is feasible and can be used for closed-loop control of continuum robots.

Data fusion of multiple kinect sensors for a rehabilitation system.

Kinect-like depth sensors have been widely used in rehabilitation systems. However, single depth sensor processes limb-blocking, data loss or data error poorly, making it less reliable. This paper focus on using two Kinect sensors and data fusion method to solve these problems. First, two Kinect sensors capture the motion data of the healthy arm of the hemiplegic patient; Second, merge the data using the method of Set-Membership-Filter (SMF); Then, mirror this motion data by the Middle-Plane; In the end, control the wearable robotic arm driving the patient's paralytic arm so that the patient can interactively and initiatively complete a variety of recovery actions prompted by computer with 3D animation games.

Drilling into cattle spines assisted by a spinal minimally invasive robot system / 中华创伤骨科杂志

Objective To assess the accuracy and safety of inserting K-wires by a spinal minimally invoive robot system into cattle spines. Methods With a program shown on the screen of a C-arm fluoroscopy system,194 nails were placed ino 17 cattle spines by a spinal minimal invasive robot system which had been developed jointly by Shenyang Institute of Autmation Chinese Academy of Sciences and our institute.Radiation frequency,radiation exposure time,operation time and deviation of drills were recorded and analyzed. Results The mean deviation in the lateral projections by the spinal minimally invasivc robot system was smaller than in the anteroposterior projections.In lateral projection,the deviations within 1 mm,2 mm and 3 mm accounted respectively for 91.6％,99.5％ and 100％.In anteroposterior projection.the deviations within 1 mm,2 mm and 3 mm accounted respecrvely for 71.1％,89.6％ and 94.8％.The mean drilling time,the mean radiation frequency and the mean radiation time per hole by the robot system were respectively 420 seconds (from 330 to 577 seconds),13 times (from 8 to 21 times) and 9 seconds (from 8 to19 seconds).There was a learning curve in mastering the system.Training for a period of time made the accuracy within 2 mm in both lateral and anteropsterior films 95.6％ of the deviations in lateral films and 92.1％ of the deviations in anteroposterior films were less than one mm. Conclusions Operators of the spinal minimally invasive robot system can avoid intraoperative X-ray radiation.The flexibility and accuracy of the system can completely satisfy the clinical needs of spinal surgeons,but there must be a learning curve.