Systems and methods for surgical margin assessment using intraoperative imaging

A surgical margin assessment system, comprising:

• an intraoperative imaging device configured to acquire imaging data from a surgical site;
• storage configured to store the imaging data;
• one or more processors;
• and
• memory storing instructions that, when executed by the one or more processors, cause the one or more processors to:
• classify individual frames of the imaging data according to view category and filter or downweight frames classified as uninformative;
• process a remaining set of frames to generate one or more margin scores;
• map the one or more margin scores to a spatially localized region adjacent a resection boundary;
• generate a candidate margin output identifying the spatially localized region;
• determine a confidence value associated with the candidate margin output;
• when the confidence value is below a threshold, generate and output a reacquisition instruction identifying the spatially localized region and specifying at least one acquisition adjustment for acquiring additional imaging of the spatially localized region;
• and
• when the confidence value satisfies the threshold, render the candidate margin output on a surgical guidance interface.

A computer-implemented method for generating training data for surgical margin assessment, the method comprising:

• acquiring intraoperative imaging data from a plurality of surgical procedures;
• acquiring pathology data for tissue removed during the plurality of surgical procedures;
• correlating the pathology data with locations represented in the intraoperative imaging data using specimen orientation information and at least one of specimen-face imaging, fiducial information, or registration operations;
• assigning a correlation confidence value to a candidate labeled example;
• generating labeled examples that identify margin-positive, margin-negative, or uncertain regions;
• and
• training a neural network using the labeled examples, wherein candidate labeled examples having correlation confidence values below a threshold are excluded or given reduced training weight.

A computer-implemented method for assisting surgical margin assessment during a surgical procedure, the method comprising:

• receiving, by a computing system, intraoperative imaging data representing tissue at or adjacent a resection boundary from at least one intraoperative imaging modality;
• processing, by one or more processors of the computing system using at least one trained machine-learning model, the intraoperative imaging data to identify one or more tissue differentials associated with the resection boundary;
• generating, based on the processing, at least one score indicative of likelihood of malignancy for each of a plurality of locations represented in the intraoperative imaging data;
• localizing, based on the at least one score, a candidate margin region within the intraoperative imaging data, wherein the localizing comprises generating one or more of a saliency map, an attention map, an anomaly map, a segmentation mask, a heatmap, or a probability map corresponding to the candidate margin region;
• and
• outputting, by the computing system, surgical guidance identifying the candidate margin region on a surgical guidance interface for use in assessing a surgical margin during the surgical procedure.