Video-based automatic hand hygiene detection for operating rooms using 3D convolutional neural networks.

Hand hygiene among anesthesia personnel is important to prevent hospital-acquired infections in operating rooms; however, an efficient monitoring system remains elusive. In this study, we leverage a deep learning approach based on operating room videos to detect alcohol-based hand hygiene actions of anesthesia providers. Videos were collected over a period of four months from November, 2018 to February, 2019, at a single operating room. Additional data was simulated and added to it. The proposed algorithm utilized a two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs), sequentially. First, multi-person of the anesthesia personnel appearing in the target OR video were detected per image frame using the pre-trained 2D CNNs. Following this, each image frame detection of multi-person was linked and transmitted to a 3D CNNs to classify hand hygiene action. Optical flow was calculated and utilized as an additional input modality. Accuracy, sensitivity and specificity were evaluated hand hygiene detection. Evaluations of the binary classification of hand-hygiene actions revealed an accuracy of 0.88, a sensitivity of 0.78, a specificity of 0.93, and an area under the operating curve (AUC) of 0.91. A 3D CNN-based algorithm was developed for the detection of hand hygiene action. The deep learning approach has the potential to be applied in practical clinical scenarios providing continuous surveillance in a cost-effective way.

Fluoroscopic subtraction Eustachian tubography: initial feasibility test in a cadaver model.

OBJECTIVES: To evaluate the technical feasibility of direct Eustachian tube catheterisation and subtraction Eustachian tubography in a cadaver model. METHODS: A total of 12 separate sessions were performed on both sides of the Eustachian tube (ET) in six human cadavers. Cadavers were positioned for the submentovertical view on a fluoroscopy table. Endoscopy-guided ET selection was used in the first three cadavers, whereas fluoroscopy-guided ET selection was used in the remaining three. Eustachian tubography was performed by injecting 2 ml of contrast media through a 5-Fr catheter. We recorded the success of ET selection, number of attempts, procedure time, and tubography quality using native and subtraction images (range, 0-3). RESULTS: Both endoscopy- and fluoroscopy-guided selections were successfully performed in five of six sessions (83.3%). There were no statistically significant differences between the endoscopy- and fluoroscopy-guided procedures in terms of the number of attempts, procedure time, rate of immediate contrast leak to the middle ear cavity, and quality of tubography (p > 0.05). An excellent quality of tubography was obtained in 83.3% (10 of 12 sessions) of subtraction images and in 33.3% (4 of 12 sessions) of native images. The tubography quality score was significantly higher for the subtraction images than for the native images (p = 0.04). CONCLUSION: Subtraction Eustachian tubography using direct catheterisation seems to be technically feasible. The entire ET can be well visualised; thus, this technique can be used as a simple tool for assessment of ET function and anatomy. KEY POINTS: • Direct catheterisation of the Eustachian tube is technically feasible. • The entire Eustachian tube could be well visualised by direct Eustachian tubography. • Subtraction Eustachian tubography images have better image quality than native images. • Subtraction Eustachian tubography can provide objective assessment of ET function and anatomy.

Transnasal Placement of a Balloon-Expandable Metallic Stent: Human Cadaver Study of the Eustachian Tube.

PURPOSE: To investigate the technical feasibility of stent placement in the cartilaginous portion of the Eustachian tube (ET). MATERIALS AND METHODS: Twelve ETs of 6 cadavers were used. Two different-sized stents were placed on either the right (2.5 mm in diameter) or left (3.5 mm in diameter) side of the ET. The procedural feasibility was assessed by subtraction Eustachian tubography, computed tomography before and after the procedure, and fluoroscopic and endoscopic images. The stent location, inner luminal diameter of the stented ET, radiation dose, procedural time, and fluoroscopy time were analyzed. RESULTS: Stent placement was successful in 11 of 12 cadaveric specimens without procedure-related complications. In the 1 specimen, the balloon catheter with crimped stent was passed into the bony canal of the ET without any resistance. The distal end of the stent was located in the middle ear cavity. Stents were located within the cartilaginous portion of the ET (n = 1), the proximal tip bridging the nasopharyngeal orifice of the ET (n = 5), or the proximal end of the stent protruded from the tubal orifice (n = 5). The mean luminal diameter in the outer segment was significantly smaller than in the middle (P < .001) and inner (P < .001) segments. The mean procedure time was 128 ± 37 seconds. The mean radiation dose and fluoroscopy time of each cadaver were 3235.4 ± 864.8 cGy/cm2 and 139 ± 49 seconds, respectively. CONCLUSIONS: Stent placement of the ET under endoscopic and fluoroscopic guidance is technically feasible in a human cadaver model.

Video recognition of simple mastoidectomy using convolutional neural networks: Detection and segmentation of surgical tools and anatomical regions.

A simple mastoidectomy is used to remove inflammation of the mastoid cavity and to create a route to the skull base and middle ear. However, due to the complexity and difficulty of the simple mastoidectomy, implementing robot vision for assisted surgery is a challenge. To overcome this issue using a convolutional neural network architecture in a surgical environment, each surgical instrument and anatomical region must be distinguishable in real time. To meet this condition, we used the latest instance segmentation architecture, YOLACT. In this study, a data set comprising 5,319 extracted frames from 70 simple mastoidectomy surgery videos were used. Six surgical tools and five anatomic regions were identified for the training. The YOLACT-based model in the surgical environment was trained and evaluated for real-time object detection and semantic segmentation. Detection accuracies of surgical tools and anatomic regions were 91.2% and 56.5% in mean average precision, respectively. Additionally, the dice similarity coefficient metric for segmentation of the five anatomic regions was 48.2%. The mean frames per second of this model was 32.3, which is sufficient for real-time robotic applications.

Modelling and manufacturing of 3D-printed, patient-specific, and anthropomorphic gastric phantoms: a pilot study.

Interventional devices including intragastric balloons are widely used to treat obesity. This study aims to develop 3D-printed, patient-specific, and anthropomorphic gastric phantoms with mechanical properties similar to those of human stomach. Using computed tomography gastrography (CTG) images of three patients, gastric phantoms were modelled through shape registration to align the stomach shapes of three different phases. Shape accuracies of the original gastric models versus the 3D-printed phantoms were compared using landmark distances. The mechanical properties (elongation and tensile strength), number of silicone coatings (0, 2, and 8 times), and specimen hardness (50, 60, and 70 Shore A) of three materials (Agilus, Elastic, and Flexa) were evaluated. Registration accuracy was significantly lower between the arterial and portal phases (3.16 ± 0.80 mm) than that between the portal and delayed phases (8.92 ± 0.96 mm). The mean shape accuracy difference was less than 10 mm. The mean elongations and tensile strengths of the Agilus, Elastic, and Flexa were 264%, 145%, and 146% and 1.14, 1.59, and 2.15 MPa, respectively, and their mechanical properties differed significantly (all p < 0.05). Elongation and tensile strength assessments, CTG image registration and 3D printing resulted in highly realistic and patient-specific gastric phantoms with reasonable shape accuracies.

Convolutional Neural Network Technology in Endoscopic Imaging: Artificial Intelligence for Endoscopy.

Recently, significant improvements have been made in artificial intelligence. The artificial neural network was introduced in the 1950s. However, because of the low computing power and insufficient datasets available at that time, artificial neural networks suffered from overfitting and vanishing gradient problems for training deep networks. This concept has become more promising owing to the enhanced big data processing capability, improvement in computing power with parallel processing units, and new algorithms for deep neural networks, which are becoming increasingly successful and attracting interest in many domains, including computer vision, speech recognition, and natural language processing. Recent studies in this technology augur well for medical and healthcare applications, especially in endoscopic imaging. This paper provides perspectives on the history, development, applications, and challenges of deep-learning technology.

Radiation Inhibits Lymph Drainage in an Acquired Lymphedema Mouse Hindlimb Model.

Background: Radiation therapy has been applied to prolong the duration of lymphedema. This study aimed to evaluate the effect of radiation on the development of lymphedema in a mouse hindlimb model. Methods and Results: A total of 24 Balb/c mice underwent the right popliteal lymph node excision and the afferent and efferent lymphatics blockage. The radiation group (n = 12) received a single 20 Gy radiation 1 day before surgery in the right hindlimb of each mouse, whereas the control group (n = 12) only received surgery without radiation. The right hindpaw thickness of each mouse was measured twice a week for 4 weeks. Fluorescence microscopy images using fluorescein isothiocyanate-dextran tracer were obtained once weekly. Immunohistochemical (IHC) staining images using anti-lymphatic vessel endothelial hyaluronan receptor-1 (anti-LYVE-1) were obtained at 4 weeks after surgery. The radiation group showed significant increase in the thickness of the right hind paws from 0.5 to 2 weeks compared with the control group. As for fluorescence lymphography, the radiation group showed a lower number of regenerated lymphatics and more congestion of tracers in the operated limb at the surgery sites at 1, 2, 3, and 4 weeks after surgery. For the IHC analysis, the radiation group showed a lower number of regenerated lymphatics per high-power field at the surgery site than the control group. Conclusion: Radiation therapy transiently aggravated the extent of lymphedema by inhibiting regenerated lymphatics in a mouse hindlimb model. However, it did not prolong the duration of lymphedema because the cutaneous interstitial flow contributes to the lymphatic fluid clearance.

Fluoroscopy-Guided Peroral Placement of a Self-Expandable Metallic Stent for Malignant Jejunal Obstruction in a Non-surgically Altered Stomach.

Malignant small bowel obstruction is a common and distressing complication in advanced cancer patients. Recently, stent placement was reported to be a safe and effective alternative treatment. However, there are only a few case reports associated with stent placement in malignant jejunal obstruction. Furthermore, most patients had a history of gastrectomy before stent placement, which shortens the catheterization pathway. In our case series, we present five cases of malignant proximal jejunal obstruction in a non-surgically altered stomach in the management of fluoroscopy-guided self-expandable metallic stent placement and discuss the interventional management and clinical outcomes. LEVEL OF EVIDENCE: Level 4, Case Series.

Development and Evaluation of a New In Vivo Volume Measuring System in Mouse Tail Lymphedema Model.

Backgrounds: Secondary lymphedema is a common complication of parasitization and breast or gynecologic cancer therapy; however, options for the treatment of lymphedema are ineffective and limited. A mouse tail model is one of the most successful animal models for a lymphatic study. Lymphedema of the mouse tail is characterized by increases in the volume of the extremity caused by accumulation of tissue fluid, proliferation of fibroblasts and adipocytes, and excessive production of collagen. However, the study of lymphedema using mouse has been plagued with difficulty in directly assessing physiologic changes owing to limitations in the measurement of the mouse tail volume. Furthermore, the mouse tail volume cannot be obtained using the general in vivo measurement method such as volumetric water displacement. Methods and Results: Lymphatic researchers have used the truncated cone formula to approximate the volume as used in the numerical approximation of a cylindrical figure. Although this formula is simple and easy to use, it has difficulties of repeatability and accuracy because the measurement procedure is highly subjective and the accuracy depends on the number of divided segments on the tail. In this article, two novel volumetric measurement methods for the mouse tail model were introduced. The methods were evaluated and compared using three mice with surgically created lymphedema on the tails. Conclusions: The two continuous measuring methods showed a possibility to improve the conventional method by continuous measurement using visual and physical detecting methods. The proposed methods facilitate the extraction of longitudinal section-specific information, which can be an important clue in a lymphatic study.

Wearable Lymphedema Massaging Modules: Proof of Concept using Origami-inspired Soft Fabric Pneumatic Actuators.

Lymphedema is a non-curative chronic swelling caused by impairment of the lymphatic system, affecting up to 250 million patients worldwide. The patients suffer from low quality of life because of discomfort and reduced range of motion due to the swelling. Severe swellings can be immediately mediated with special massaging technique known as the Manual Lymphatic Drainage (MLD). Limitations of MLD involves long travel distances, the cost of regular treatment sessions, and the lack of lymphedema specialists. Since MLD is performed very gently, described as caressing a baby's head, soft wearable robotics with its inherent compliance and safety is the perfect solution to creating a light and safe wearable lymphedema massaging device. In this paper, origami-inspired soft fabric pneumatic actuator is developed that creates not only normal force, but also shear force which is essential in the performance of MLD. The shear is created by the unfolding of the Z-shaped fold-lines as the actuator is inflated. One Z-folded actuator module of 30 x 60 mm dimension with a single fold of 15 mm fold height creates maximum shear force of about 1.5 N and stroke displacement of about 30 mm when subjected to compression loading of 5 N. The range of forces exerted can be tuned by varying the tension of the compressive clothing covering the actuators, and the stroke displacement can be varied by changing the parameter of the actuator module itself, such as the fold height and the number of the folds. The modules can also be repeatedly actuated under compressive clothing, and therefore, the developed actuator modules have high potential as a wearable massaging device.