Search

Adequate clinical evaluation of artificial intelligence (AI) algorithms before adoption in practice is critical. Clinical evaluation aims to confirm acceptable AI performance through adequate external testing and confirm the benefits of AI-assisted care compared with conventional care through appropriately designed and conducted studies, for which prospective studies are desirable. This article explains some of the fundamental methodological points that should be considered when designing and appraising the clinical evaluation of AI algorithms for medical diagnosis. The specific topics addressed include the following

Artificial intelligence (AI) will likely affect various fields of medicine. This article aims to explain the fundamental principles of clinical validation, device approval, and insurance coverage decisions of AI algorithms for medical diagnosis and prediction. Discrimination accuracy of AI algorithms is often evaluated with the Dice similarity coefficient, sensitivity, specificity, and traditional or free-response receiver operating characteristic curves. Calibration accuracy should also be assessed, especially for algorithms that provide probabilities to users. As current AI algorithms have limited generalizability to real-world practice, clinical validation of AI should put it to proper external testing and assisting roles. External testing could adopt diagnostic case-control or diagnostic cohort designs. A diagnostic case-control study evaluates the technical validity/accuracy of AI while the latter tests the clinical validity/accuracy of AI in samples representing target patients in real-world clinical scenarios. Ultimate clinical validation of AI requires evaluations of its impact on patient outcomes, referred to as clinical utility, and for which randomized clinical trials are ideal. Device approval of AI is typically granted with proof of technical validity/accuracy and thus does not intend to directly indicate if AI is beneficial for patient care or if it improves patient outcomes. Neither can it categorically address the issue of limited generalizability of AI. After achieving device approval, it is up to medical professionals to determine if the approved AI algorithms are beneficial for real-world patient care. Insurance coverage decisions generally require a demonstration of clinical utility that the use of AI has improved patient outcomes.

This paper introduces a newly developed robotic gait training system for lower-limb rehabilitation of stroke patients. The system (Cyborg-Trainer L; Cyborg-Lab Co., Korea) provides a stair-climbing mode in addition to the conventional level-walking mode by leveraging a unique exoskeleton structure with separately operable foot plates. Unlike conventional end-effector type gait training robots, the subject’s feet are not constrained by foot plates, but are free to emulate the ground or a set of stairs. The ground reaction force is measured by force sensors in the foot plates and utilized to compensate for the vertical movement of pelvis. The exoskeleton structures are connected at hip, knee, and ankle joints, and these can support a patient’s weight to ensure a normal gait pattern. The system has four control modes with different levels of assistive or resistance force. To show the feasibility of the developed training mode, a series of experiments measuring muscle activity were conducted during 1) level-walking with the robot, 2) level-walking on a treadmill without a robot, 3) stair-climbing with the robot, and 4) actual stairclimbing without a robot. The muscle activation from the rectus femoris, biceps femoris, tibialis anterior, and gastrocnemius medialis of the dominant leg of five healthy adults were measured and analyzed. Results showed that all muscles had a rhythmic muscle activation pattern. Even though muscle activation patterns were different between gaits using the robotic gait system and those not using it, reduced amplitudes and phasic muscle activations were observed during the training in the robotic system. The developed system is a new type of robotic gait training system that could induce phasic lower limb muscle activation patterns, and its clinical efficacy will be validated in clinical trials after regulatory approval.

In this paper we present a new robotic device for performing a percutaneous needle biopsy. The device is designed to perform introducer insertion and tissue sampling in organs of interest, such as the lung and liver. It is designed to have smaller parts and better performance than the previous models, and includes light-weight, sensor-based initialization, needle insertion force monitoring, and high performance tissue sampling. With some design improvements, such as a small pillar gear, for needle cartridge rotation, thin side frames, and a kinematic, linkage-based, spring block trigger, the device is over 40% lighter than previous models. Hall sensors mounted on each moving part, and a load cell between a needle insertion part and the frame, provide information for the stable and safe operation of the device. A prototype was fabricated, based on the design model, and the performance was evaluated for the positioning of the part holding the needle and for the tissue sampling. The results show that the device has an average error of less than 0.1 mm and a tissue sampling success rate of 100%, in both ex-vivo and in-vivo experiments.

To reduce the radiation exposure of personnel during an interventional procedure for arrhythmia, a robot has been developed and implemented herein for use in interventional procedures. Studies on the control of an electrophysiology catheter by robots are being conducted. However, controlling a catheter using a robot has limited precision owing to external forces subjected on the catheter due to blood flow and pulse inside a heart. This study implements a reinforcement learning method for automated control of a catheter by a robot. Using the reinforcement learning method, this study aims to show that such a robot can learn to manipulate a catheter to reach a target in a simulated environment and subsequently control a catheter in an actual environment. Randomization noise is used during the simulation to reduce the differences between the simulation and actual learning environments. Each environment is implemented with different movement values depending on insertion angles and steps of the catheter model. When the results from the simulated learning model are implemented in the actual environment, the success rate of catheter reaching the designated target is 73 %. In addition, the noise-implemented model shows that the success rate can be increased up to 87 %. Through these experiments, the study verifies that a simulated learning model can be implemented in a robot system to control an actual catheter as well as that the success rate of the model can be increased using randomization noise.

BACKGROUND: In vitro generation of three-dimensional vessel network is crucial to investigate and possibly improve vascularization after implantation in vivo. This work has the purpose of engineering complex tissue regeneration of a vascular network including multiple cell-type, an extracellular matrix, and perfusability for clinical application. METHODS: The two electrospun membranes bonded with the vascular network shape are cultured with endothelial cells and medium flow through the engineered vascular network. The flexible membranes are bonded by amine-epoxy reaction and examined the perfusability with fluorescent beads. Also, the perfusion culture for 7 days of the endothelial cells is compared with static culture on the engineered vascular network membrane. RESULTS: The engineered membranes are showed perfusability through the vascular network, and the perfused network resulted in more cell proliferation and variation of the shear stress-related genes expression compared to the static culture. Also, for the generation of the complex vascularized network, pericytes are co-cultured with the engineered vascular network, which results in the Collagen I is expressed on the outer surface of the engineered structure. CONCLUSION: This study is showing the perfusable in vitro engineered vascular network with electrospun membrane. In further, the 3D vascularized network module can be expected as a platform for drug screening and regenerative medicine.

Conventional vascular intervention procedures present issues including X-ray exposure during operation, and an experience-dependent success rate and clinical outcome. This paper presents a novel robotic system using modularized bi-motional roller cartridge assemblies for robotic vascular interventions, specifically percutaneous coronary interventions (PCIs). The patient-side robot manipulates instruments such as the guiding catheter, guidewire, balloon/stent catheter, and diagnostic sensor catheter via commands from the user interface device, which is controlled by the physician. The proposed roller cartridge assembly can accommodate instruments of various sizes with an active clamping mechanism, and implements simultaneous translation and rotation motions. It also implements force feedback in the physician-side system, to effectively monitor the patient-side system’s status. The positioning accuracy and precision in using the robotic system showed satisfactory performance in a phantom-based test. It was also confirmed, through animal experiments and a pilot clinical trial, that the system demonstrates feasibility for clinical use.

Skeletal muscle mass deficiency and quality degradation constitute sarcopenia for elderly people. Sarcopenia can result in musculoskeletal damage and accompany various metabolic problems, which make early sarcopenia diagnosis important. Various modalities, such as computed tomography (CT) and magnetic resonance imaging (MRI), have been developed for screening sarcopenia. Recently, ultrasound scanning was suggested for screening sarcopenia because of its safety, usability, and cost effectiveness. However, there has been no standardized assessment methodology for screening sarcopenia with ultrasound scanning. Therefore, prior to this study, we developed a four-degrees-of-freedom (DOF) sarcopenia detection system using an RGB-D camera and an ultrasound probe to automatically scan the human thigh without operator dependency. However, due to the eye-to-hand approach with the RGB-D camera, the system has limited usability for clinical trials. Therefore, in this study we modified the system such that it became eye-in-hand by attaching the RGB-D camera to the upper part of the system with an enhanced arc fitting algorithm. The modified system and enhanced algorithm were verified by an in-vitro test with bean curd-gelatin phantom. The results showed that the thickness of bean curd in the gelatin phantom was maintained at approximately 12.7 ± 0.35 mm over the 71.5∘ scanning range with 2.49 ± 0.15 N radial force at various thickness measuring points.

This paper presents a novel patent pending 4-to-1 switched-capacitor converter (SCC) for fixed ratio step down conversion applications. Optimization method is introduced to get the maximum efficiency at target output current. This shows the optimized region of operation that set up a performance comparison framework. Analysis shows that the proposed topology has performance advantages over the other four popular topologies. The proposed work is designed in 130nm technology and achieves 96.7% efficiency at 8A output current. Two 16-to-4V SCC prototypes for smart phone application using discrete switches are built and tested. Doubler cascade (existing commercial solution) shows 95.1% peak efficiency (including gate driving loss) and 93.2% at 12A output current. The proposed topology achieves 95.8% peak efficiency (including gate driving loss) and 94.1% at 12A output current. The proposed topology can be generalized to any exponential ratio of conversion like doubler topology in multiple cascade stages.

The purpose of this study was to demonstrate self-organizing in vitro multicellular tumor spheroid (MCTS) formation in a microfluidic system and to observe the behavior of MCTSs under controlled microenvironment. The employed microfluidic system was designed for simple and effective formation of MCTSs by generating nutrient and oxygen gradients. The MCTSs were composed of cancer cells, vascular endothelial cells, and type I collagen matrix to mimic the in vivo tumor microenvironment (TME). Cell culture medium was perfused to the microfluidic device loaded with MCTSs by a passive fluidic pump at a constant flow rate. The dose response to an MMPs inhibitor was investigated to demonstrate the effects of biochemical substances. The result of long-term stability of MCTSs revealed that continuous perfusion of cell culture medium is one of the major factors for the successful MCTS formation. A continuous flow of cell culture medium in the in vitro TME greatly affected both the proliferation of cancer cells in the micro-wells and the sustainability of the endothelial cell-layer integrity in the lumen of microfluidic channels. Addition of MMP inhibitor to the cell culture medium improved the stability of the collagen matrix by preventing the detachment and shrinkage of the collagen matrix surrounding the MCTSs. In summary, the present constant flow assisted microfluidic system is highly advantageous for long-term observation of the MCTS generation, tumorous tissue formation process and drug responses. MCTS formation in a microfluidic system may serve as a potent tool for studying drug screening, tumorigenesis and metastasis.

X-ray angiography, used in the evaluation of coronary artery disease, presents difficulties in the performance of quantitative coronary angiography analysis, by identifying major vessels. These difficulties are due to problems such as non-uniform illumination, low contrast ratio, and the presence of other tissues. Therefore, segmentation of the desired vessels in images containing multiple blood vessels is clinically important. This paper proposes selective feature mapping as a method for segmenting the left anterior descending artery main vessel in coronary angiography images. The proposed method consists of two steps for generating a candidate area of an image and then segmenting it. To generate the candidate area, feature maps that overlap significantly with the area of the ground truth are selected and combined. Segmentation then is performed using a neural network that learns only the ground truth region of the input image. The proposed method consists of eight modules: pre-processing of the angiogram, resizing of ground truth, pre-processing for segmentation, post-processing for segmentation, network, and segmentation network. This method has a precision of about 0.066, recall of 0.091, and an F1 score of 0.094, values which are higher than those generated by the U-Net, one of the conventional techniques.

This paper presents a new master manipulator applied to robotic systems for arrhythmia ablation. The manipulator is designed to implement the concept of two different master-slave teleoperation controls in the clinical application. One is the conventional control between the master and catheter-handle in the slave site and the other is catheter tip manipulation corresponding the master motion. For the purpose, the master has six degrees of freedom (DOF) and consists of three main components: a spherical mechanism for rotational motion of 2-DOF, 3-RRR planar parallel mechanism with 3-DOF, and counter-weight lifting mechanism for the vertical movement. Two mechanisms except the lifting mechanism are parallel chains and structurally more complicated. Therefore, their forward kinematics are analyzed, and workspaces are simulated. To evaluate the applicability for robotic catheter systems, the manipulator prototype was tested for its smoothness, workspace, and teleoperation performance. The subjects in the smoothness test reported no considerable friction and jerk in free movement in the three-dimensional space as shown in the recorded curves. The workspace test shows the actual workspace is similar to the simulated one except some structural constraints in the 3-RRR mechanism. For the last test, a robotic catheter teleoperation system with a slave robot and the software structure for the robotic catheter control system with electrical connection of each component were built. The result performed by five different users shows the motion of the slave robot well tracks that of the master device with small average errors and time delay that are acceptable in robotic catheter teleoperation systems.

This paper proposes a catheter robot system for cardiac ablation therapy. The robot mechanism for actuating the catheter and the control methods of the catheter tip position will be suggested. The robot system consists of a front module and a handle-control module. The front module is a dual-gripper mechanism for inserting and rotating the catheter without slip between the catheter and gripper, and the handle-control module is for controlling the knob for the bending motion of the catheter. The proposed dual-gripper mechanism has the advantage of large contact surface between the grippers and the catheter, like the human finger, allowing it to form a stable grip of the catheter. However, for continuous insertion, the grippers should be controlled interactively. To achieve this, we explain the motion-planning and trajectory-generation methods. Moreover, we propose a control method that can intuitively control the position of the catheter using a mapping system as manual work. We explain the method of updating the Jacobian model in real time using two position sensors attached to the catheter. Finally, we show the simulation and experimental results to confirm the proposed mechanism and control methods useful for controlling the catheter

Sarcopenia is defined as muscle mass and strength loss with aging. As places, such as South Korea, Japan, and Europe have entered an aged society, sarcopenia is attracting global attention with elderly health. However, only few developed devices can quantify sarcopenia diagnosis modalities. Thus, the authors developed a sarcopenia detection system with 4 degrees of freedom to scan the human thigh with ultrasound probe and determine whether he/she has sarcopenia by inspecting the length of muscle thickness in the thigh by ultrasound image. To accurately measure the muscle thickness, the ultrasound probe attached to the sarcopenia detection system, must be moved angularly along the convex surface of the thigh with predefined pressure maintained. Therefore, the authors proposed an angular thigh scanning method for the aforementioned reason. The method first curve-fits the angular surface of the subject&rsquo, s thigh with piecewise arcs using D information from a fixed RGB-D camera. Then, it incorporates a Jacobian-based ultrasound probe moving method to move the ultrasound probe along the curve-fitted arc and maintains radial interface force between the probe and the surface by force feedback control. The proposed method was validated by in-vitro test with a human thigh mimicked ham-gelatin phantom. The result showed the ham tissue thickness was maintained within approximately 26.01 &plusmn, 1.0 mm during 82∘ scanning with a 2.5 N radial force setting and the radial force between probe and surface of the phantom was maintained within 2.50 &plusmn, 0.1 N.

BACKGROUND Endoscopes are widely used for visualizing the respiratory tract, urinary tract, uterus, and gastrointestinal tracts. Despite high demand, people in underdeveloped and developing countries cannot obtain proper access to endoscopy. Moreover, commercially available endoscopes are mostly nonarticulable as well as not actively controlled, limiting their use. Articulating endoscopes are required for some diagnosis procedures, due to their ability to image wide areas of internal organs. Furthermore, actively controlled articulating endoscopes are less likely to harm the lumen than rigid endoscopes because they can avoid contact with endothelial tissues. OBJECTIVE The study aimed to demonstrate the feasibility and acceptability of smartphone-based wide-field articulable endoscope system for minimally invasive clinical applications in developing and less developed countries. METHODS A thin articulable endoscope system that can be attached to and actively controlled by a smartphone was designed and constructed. The system consists of a flexible endoscopic probe with a continuum mechanism, 4 motor modules for articulation, a microprocessor for controlling the motor with a smartphone, and a homebuilt app for streaming, capturing, adjusting images and video, and controlling the motor module with a joystick-like user interface. The smartphone and motor module are connected via an integrated C-type On-The-Go (OTG) USB hub. RESULTS We tested the device in several human-organ phantoms to evaluate the usability and utility of the smartphone-based articulating endoscope system. The resolution (960 × 720 pixels) of the device was found to be acceptable for medical diagnosis. The maximum bending angle of 110° was designed. The distance from the base of the articulating module to the tip of the endoscope was 45 mm. The angle of the virtual arc was 40.0°, for a curvature of 0.013. The finest articulation resolution was 8.9°. The articulating module succeeded in imaging all 8 octants of a spherical target, as well as all 4 quadrants of the indices marked in human phantoms. CONCLUSIONS The portable wide-field endoscope was successfully controlled using a smartphone, yielding clear images with a resolution of 960 × 720 pixels at realistic focal distances. Actively and precisely controlled articulating movements have resulted in minimally invasive monitoring in the narrow space of internal organs providing a wide-area view. We found our smartphone-based active articulated endoscope to be suitable for point-of-care applications in developing and less developed countries.

More than half of post-operative complications can be prevented, and operation performances can be improved based on the feedback gathered from operations or notifications of the risks during operations in real time. However, existing surgical analysis methods are limited, because they involve time-consuming processes and subjective opinions. Therefore, the detection of surgical instruments is necessary for (a) conducting objective analyses, or (b) providing risk notifications associated with a surgical procedure in real time. We propose a new real-time detection algorithm for detection of surgical instruments using convolutional neural networks (CNNs). This algorithm is based on an object detection system YOLO9000 and ensures continuity of detection of the surgical tools in successive imaging frames based on motion vector prediction. This method exhibits a constant performance irrespective of a surgical instrument class, while the mean average precision (mAP) of all the tools is 84.7, with a speed of 38 frames per second (FPS).

In this paper, we propose a new technique to recognize vessels in robot-assisted laparoscopic surgery images by using surgical instruments. The proposed method does not require additional hardware or parameter adjustment because it detects blood vessels by using only the color information of the image. The concept of a hessian matrix is used in the HSV color space of the image to detect the edges of the blood vessels. In addition, the histogram equalization technique, clustering technique, and region growing are used to remove the surgical tools. Images of actual robot-assisted laparoscopic surgery videos were used. The processing speed was approximately 0.3 s per frame at 640p and approximately 0.8 s per frame at 1280p. The average recall was 92.67%.

The aim of this study was to develop a multi-nozzle based bioprinting system for fabrication of three-dimensional (3D) biological structure. In this study, a thermoplastic biomaterial that has relatively high mechanical stability, polycaprolactone (PCL) was used to make the 3D structure. A multi-nozzle bioprinting system was designed to dispense thermoplastic biomaterial and hydrogel simultaneously. The system that consists of 3-axes of x-y-z motion control stage and a compartment for injection syringe control mounted on the stage has been developed. Also, it has 1-axis actuator for position change of nozzle. The controllability of the printed line width with PCL was tested as a representative performance index.

This paper presents a novel end-effector designed for a teleoperation type robotic system to perform automated sampling for needle biopsies. The objectives of the robotic system are to realize teleoperation control of the pose and position of the biopsy instrument to reduce X-ray exposure and enhance the precision and accuracy of the procedure as well as to automate the sequence of multiple samplings during the biopsy. In order to realize the desired functions, the end-effector comprises three modules: the needle cartridge, needle positioning mechanism, and needle driving unit. The detachable needle cartridge contains six needle sets: one guide needle and five sampling needles. It rotates itself to change needles and rotates the needle body to adjust the facing angle of the needle tip. The needle positioning mechanism is spherical with two degrees of freedom and is a modification of the typical hemisphere type. The needle driving unit controls the needle grab/release and insertion/retreat motions. A concept design and prototype were developed with an experimental master-salve system for the teleoperation. The experimental results successfully demonstrated the feasibility of the robotic biopsy procedure.

Although the use of the surgical robot is rapidly expanding for various medical treatments, there still exist safety issues and concerns about robot-assisted surgeries due to limited vision through a laparoscope, which may cause compromised situation awareness and surgical errors requiring rapid emergency conversion to open surgery. To assist surgeon’s situation awareness and preventive emergency response, this study proposes situation information guidance through a vision-based common algorithm architecture for automatic detection and tracking of intraoperative hemorrhage and surgical instruments. The proposed common architecture comprises the location of the object of interest using feature texture, morphological information, and the tracking of the object based on Kalman filter for robustness with reduced error. The average recall and precision of the instrument detection in four prostate surgery videos were 96% and 86%, and the accuracy of the hemorrhage detection in two prostate surgery videos was 98%. Results demonstrate the robustness of the automatic intraoperative object detection and tracking which can be used to enhance the surgeon’s preventive state recognition during robot-assisted surgery.

This paper focuses on international standards and guidelines related to evaluating the safety and performance of wearable dialysis systems and devices. The applicable standard and evaluation indices for safety and performance are determined, and the relevant international standards and guidelines are provided in a table. In addition, example experiments using a triaxial accelerometer and robot arm are presented for testing the endurance and safety of wearable artificial kidneys. The findings in this paper can be used to suggest new guidelines for the mechanical safety and performance evaluation of wearable artificial kidney systems.

This paper proposes intelligent fuzzy proportional–integral–derivative (PID) controllers for the position control of a master–slave configuration laparoscopic surgery robot system. For the slave robot controller, two fuzzy PID control algorithms are implemented: a rule-based fuzzy control algorithm for online PID gain tuning and a learning fuzzy controller to tune the rules in the rule buffer automatically online. The two fuzzy controllers are tested using sinusoidal reference motions with various periods and the test results are compared with those of a conventional PID controller in terms of position control performance. Various performance indices are used for the comparison, including root mean square error, steady-state error, integral of average error, integral squared error, integral time-weighted absolute error, and integral of time multiplied by the squared error. The evaluation shows that the learning fuzzy controller yields the best performance among the three algorithms. Furthermore, the relationship between the scaling gains and the performance can be deduced to construct a comparative tuning algorithm, which enables the scaling gains to be optimally tuned with less trial and error. Further refinement of the algorithm for enhancing control performance is under way.