Abstract: Embodiments described herein provide various examples of monitoring adverse events in the background while displaying a higher-resolution surgical video on a lower-resolution display device. In one aspect, a process for detecting adverse events during a surgical procedure can begin by receiving a surgical video. The process then displays a first portion of the video images of the surgical video on a screen to assist a surgeon performing the surgical procedure. While displaying the first portion of the video images, the process uses a set of deep-learning models to monitor a second portion of the video images not being displayed on the screen, wherein each deep-learning model is constructed to detect a given adverse event among a set of adverse events. In response to detecting an adverse event in the second portion of the video images, the process notifies the surgeon of the detected adverse event to prompt an appropriate action.

Abstract: A system and method are provided for identifying a multimodal biomarker of a prognostic prediction, using a deep learning framework trained to analyze different modality data, including radiomic image data, pathology image data, and molecular image data to obtain unimodal embedding predictions from those modality data and generate multimodal embedding predictions, through application of a loss minimization and attention-based fusion processes.

Abstract: Embodiments described herein provide various examples of displaying an endoscope video of an endoscope resolution on a display having a screen resolution lower than the endoscope resolution. In one aspect, a process of displaying the endoscope video on the display can begin by displaying a portion of the endoscope video having the same resolution as the screen resolution on the display. While displaying the portion of the endoscope video, the process detects a surgical tool present in an off-screen portion of the endoscope video not visible on the display, and therefore not visible to the user. The process further analyzes the off-screen portion of the endoscope video including the detected surgical tool. The process subsequently generates an alert when a risk of the detected surgical tool is identified.

Abstract: An endoscope system capable of automatically turning on/off a light source during a surgical procedure is disclosed. This endoscope system includes: an endoscope module; a light source module coupled to the endoscope module; and a light-source control module. Moreover, the light-source control module controls an ON/OFF state of the light source by: (1) receiving real-time video images captured by the endoscope module; (2) processing the real-time video images to determine whether the endoscope module is inserted into a patient's body or is outside of the patient's body; and (3) in response to determining the endoscope module being outside of the patient's body while the light source is turned on, generating a control signal to immediately turn off the light source to the endoscope module, thereby ensuring safety of people in the operating room and preventing sensitive information in the operating room from being captured by the endoscope module.

Abstract: This disclosure provides techniques of synchronizing the playback of two recorded videos of the same surgical procedure. In one aspect, a process for generating a composite video from two recorded videos of a surgical procedure is disclosed. This process begins by receiving a first and second surgical videos of the same surgical procedure. The process then performs phase segmentation on each of the first and second surgical videos to segment the first and second surgical videos into a first set of video segments and a second set of video segments, respectively, corresponding to a sequence of predefined phases. Next, the process time-aligns each video segment of a given predefined phase in the first video with a corresponding video segment of the given predefined phase in the second video. The process next displays the time-aligned first and second surgical videos for comparative viewing.

Abstract: Embodiments described herein provide various examples of a surgical video analysis system for segmenting surgical videos of a given surgical procedure into shorter video segments and labeling/tagging these video segments with multiple categories of machine learning descriptors. In one aspect, a process for processing surgical videos recorded during performed surgeries of a surgical procedure includes the steps of: receiving a diverse set of surgical videos associated with the surgical procedure; receiving a set of predefined phases for the surgical procedure and a set of machine learning descriptors identified for each predefined phase in the set of predefined phases; for each received surgical video, segmenting the surgical video into a set of video segments based on the set of predefined phases and for each segment of the surgical video of a given predefined phase, annotating the video segment with a corresponding set of machine learning descriptors for the given predefined phase.

Abstract: In this patent disclosure, a machine-learning-based system for automatically turning on/off a light source of an endoscope camera during a surgical procedure to ensure the safety of the surgical staff is disclosed. The disclosed system can receive a real-time video image captured by the endoscope camera, wherein the real-time video image is captured either inside the patient's body or outside of the patient's body. The system next processes the real-time video image using a first statistical classifier to classify the real-time video image as either being inside the patient's body or being outside of the patient's body. If the real-time video image is classified as being outside of the patient's body, the system next determines if the light source is turned on. If so, the system generates a control signal to immediately turn off the light source. Otherwise, the system continues receiving and processing real-time video images captured by the endoscope camera.

Abstract: Embodiments described herein provide various examples of synchronizing the playback of a recorded video of a surgical procedure with a live video feed of a user performing the surgical procedure. In one aspect, a system can simultaneously receive a recorded video of a surgical procedure and a live video feed of a user performing the surgical procedure in a training session. More specifically, the recorded video is shown to the user as a training reference, and the surgical procedure includes a set of surgical tasks. The system next simultaneously monitors the playback of a current surgical task in the set of surgical tasks in the recorded video and the live video feed depicting the user performing the current surgical task. Next, the system detects that the end of the current surgical task has been reached during the playback of the recorded video.

Abstract: This patent disclosure provides various embodiments of combining multiple modalities of non-text surgical data in forms of videos, images, and audios in a meaningful manner so that the combined data can be used to perform comprehensive data analytics for a surgical procedure. In some embodiments, the disclosed system can begin by receiving two or more modalities of surgical data during the surgical procedure. The system then time-synchronizes the two or more modalities of surgical data to generate two or more modalities of time-synchronized surgical data. Next, the system converts each modality of the time-synchronized surgical data into a corresponding array of values of a common format. The system then combines the two or more arrays of values to generate a combined set of values. The system subsequently performs comprehensive data analytics on the combined set of values to generate a surgical decision for the surgical procedure.

Abstract: Embodiments described herein provide various examples of a surgical video analysis system for segmenting surgical videos of a given surgical procedure into shorter video segments and labeling/tagging these video segments with multiple categories of machine learning descriptors. In one aspect, a process for processing surgical videos recorded during performed surgeries of a surgical procedure includes the steps of: receiving a diverse set of surgical videos associated with the surgical procedure; receiving a set of predefined phases for the surgical procedure and a set of machine learning descriptors identified for each predefined phase in the set of predefined phases; for each received surgical video, segmenting the surgical video into a set of video segments based on the set of predefined phases and for each segment of the surgical video of a given predefined phase, annotating the video segment with a corresponding set of machine learning descriptors for the given predefined phase.

Abstract: Embodiments described herein provide various examples of mitigating electrical burn risks when using a monopolar electrosurgery tool on a patient in an electrosurgery procedure. In one aspect, a process receives an electrode configuration of the monopolar electrosurgery tool which includes a location of an active electrode of the monopolar electrosurgery tool at a surgical site on the patient's body and a location of a return electrode of the monopolar electrosurgery tool elsewhere on the patient's body. The process further obtains information of a metal implant inside the patient's body. Next, the process computes a plurality of potential current paths between the locations of the active electrode and the return electrode. The process then determines if one or more current paths in the plurality of computed potential current paths flow through the metal implant. If so, the process informs a surgical staff of potential electrical burn injuries associated with the electrode configuration.

Abstract: This patent disclosure provides various verification techniques to ensure that anonymized surgical procedure videos are indeed free of any personally-identifiable information (PII). In a particular aspect, a process for verifying that an anonymized surgical procedure video is free of PII is disclosed. This process can begin by receiving a surgical video corresponding to a surgery. The process next removes personally-identifiable information (PII) from the surgical video to generate an anonymized surgical video. Next, the process selects a set of verification video segments from the anonymized surgical procedure video. The process subsequently determines whether each segment in the set of verification video segments is free of PII. If so, the process replaces the surgical video with the anonymized surgical video for storage. If not, the process performs additional PII removal steps on the anonymized surgical video to generate an updated anonymized surgical procedure video.

Abstract: Embodiments described herein provide various examples of a machine-learning-based visual-haptic system for constructing visual-haptic models for various interactions between surgical tools and tissues. In one aspect, a process for constructing a visual-haptic model is disclosed. This process can begin by receiving a set of training videos. The process then processes each training video in the set of training videos to extract one or more video segments that depict a target tool-tissue interaction from the training video, wherein the target tool-tissue interaction involves exerting a force by one or more surgical tools on a tissue. Next, for each video segment in the set of video segments, the process annotates each video image in the video segment with a set of force levels predefined for the target tool-tissue interaction. The process subsequently trains a machine-learning model using the annotated video images to obtain a trained machine-learning model for the target tool-tissue interaction.

Abstract: This patent disclosure provides various embodiments of combining multiple modalities of non-text surgical data of different formats, in particular in forms of videos, images, and audios in a meaningful manner so that the combined data from the multiple modalities are compatible with text data. In some embodiments, prior to combining the multiple modalities of surgical data, multiple segmentation engines are used to segment and convert a corresponding modality of surgical data into a corresponding set of metrics and parameters. The multiple sets of metrics and parameters corresponding to the multiple modalities are then combined to generate a combined feature set. The combined feature set can be provided to a data analytics tool for performing comprehensive data analyses on the combined feature set to generate one or more predictions for the surgical procedure.

Abstract: Embodiments described herein provide various examples of displaying an endoscope video of an endoscope resolution on a display having a screen resolution lower than the endoscope resolution. In one aspect, a process of displaying the endoscope video on the display can begin by displaying a portion of the endoscope video having the same resolution as the screen resolution on the display. While displaying the portion of the endoscope video, the process detects a surgical tool present in an off-screen portion of the endoscope video not visible on the display, and therefore not visible to the user. The process further analyzes the off-screen portion of the endoscope video including the detected surgical tool. The process subsequently generates an alert when a risk of the detected surgical tool is identified.

Abstract: Embodiments described herein provide various examples of predicting potential current paths from an active electrode of a monopolar electrosurgery tool to a return electrode of the monopolar electrosurgery tool based on analyzing electrical properties of tissues inside a patient's body, and evaluating and eliminating tissue burn risks associated with the predicted current paths. In some embodiments, a current-path-prediction technique is used to predict a set of potential current paths from the active electrode to the return electrode for any given geometrical configuration of the two electrodes on the patient's body. These predicted current paths can then be pictorially displayed on a 3D scan of the patient's body or an endoscopic view of the patient's body and in relation to the display of any existing metal implant inside the patient's body, which allows for visualizing points of tissue burn risks inside the patient's body.

Abstract: This patent disclosure provides various embodiments for anonymizing raw surgical procedure videos recorded by a recording device, such as an endoscope camera, during a surgical procedure performed on a patient inside an operating room (OR). In one aspect, a process for anonymizing raw surgical procedure videos recorded by a recording device within an OR is disclosed. This process can begin by receiving a set of raw surgical videos corresponding to a surgical procedure performed within the OR. The process next merges the set of raw surgical videos to generate a surgical procedure video corresponding to the surgical procedure. Next, the process detects image-based personally-identifiable information embedded in the set of raw video images of the surgical procedure video. When image-based personally-identifiable information is detected, the process automatically de-identifies the detected image-based personally-identifiable information in the surgical procedure video.

Abstract: Embodiments described herein provide various examples of a visual-haptic feedback system for generating a haptic feedback signal based on captured endoscopy images. In one aspect, the process for generating the haptic feedback signal includes the steps of: receiving an endoscopic video captured for a surgical procedure performed on a robotic surgical system; detecting a surgical task in the endoscopic video involving a given type of surgical tool-tissue interaction; selecting, a machine learning model constructed for analyzing the given type of surgical tool-tissue interaction; for a video image associated with the detected surgical task depicting the given type of surgical tool-tissue interaction, applying the selected machine learning model to the video image to predict a strength level of the depicted surgical tool-tissue interaction; and then providing the predicted strength level to a surgeon performing the surgical task as a haptic feedback signal for the given type of surgical tool-tissue interaction.

Abstract: Embodiments described herein provide various examples of displaying video images of a surgical video captured at a first resolution on a screen of a surgical system having a second resolution lower than the first resolution. In one aspect, a process begins by receiving the surgical video and selecting a first portion of the video images having the same or substantially the same resolution as the second resolution. The process subsequently displays the first portion of the video images on the screen. While displaying the first portion of the video images, the process monitors a second portion of the video images not being displayed on the screen for a set of predetermined events, wherein the second portion is not visible to the user. When a predetermined event in the set of predetermined events is detected in the second portion, the process generates an alert to notify the user.

Abstract: Embodiments described herein provide various examples of a system for extracting an actual procedure duration composed of actual surgical tool-tissue interactions from an overall procedure duration of a surgical procedure on a patient. In one aspect, the system is configured to obtain the actual procedure duration by: obtaining an overall procedure duration of the surgical procedure; receiving a set of operating room (OR) data from a set of OR data sources collected during the surgical procedure, wherein the set of OR data includes an endoscope video captured during the surgical procedure; analyzing the set of OR data to detect a set of non-surgical events during the surgical procedure that do not involve surgical tool-tissue interactions; extracting a set of durations corresponding to the set of non-surgical events; and determining the actual procedure duration by subtracting the set of extracted durations from the overall procedure duration.