Comparative analysis of the spatial distribution of brain metastases across several primary cancers using machine learning and deep learning models.

OBJECTIVE: Brain metastases (BM) are associated with poor prognosis and increased mortality rates, making them a significant clinical challenge. Studying BMs can aid in improving early detection and monitoring. Systematic comparisons of anatomical distributions of BM from different primary cancers, however, remain largely unavailable. METHODS: To test the hypothesis that anatomical BM distributions differ based on primary cancer type, we analyze the spatial coordinates of BMs for five different primary cancer types along principal component (PC) axes. The dataset includes 3949 intracranial metastases, labeled by primary cancer types and with six features. We employ PC coordinates to highlight the distinctions between various cancer types. We utilized different Machine Learning (ML) algorithms (RF, SVM, TabNet DL) models to establish the relationship between primary cancer diagnosis, spatial coordinates of BMs, age, and target volume. RESULTS: Our findings revealed that PC1 aligns most with the Y axis, followed by the Z axis, and has minimal correlation with the X axis. Based on PC1 versus PC2 plots, we identified notable differences in anatomical spreading patterns between Breast and Lung cancer, as well as Breast and Renal cancer. In contrast, Renal and Lung cancer, as well as Lung and Melanoma, showed similar patterns. Our ML and DL results demonstrated high accuracy in distinguishing BM distribution for different primary cancers, with the SVM algorithm achieving 97% accuracy using a polynomial kernel and TabNet achieving 96%. The RF algorithm ranked PC1 as the most important discriminating feature. CONCLUSIONS: In summary, our results support accurate multiclass ML classification regarding brain metastases distribution.

Simulated outcomes for durotomy repair in minimally invasive spine surgery.

Minimally invasive spine surgery (MISS) is increasingly performed using endoscopic and microscopic visualization, and the captured video can be used for surgical education and development of predictive artificial intelligence (AI) models. Video datasets depicting adverse event management are also valuable, as predictive models not exposed to adverse events may exhibit poor performance when these occur. Given that no dedicated spine surgery video datasets for AI model development are publicly available, we introduce Simulated Outcomes for Durotomy Repair in Minimally Invasive Spine Surgery (SOSpine). A validated MISS cadaveric dural repair simulator was used to educate neurosurgery residents, and surgical microscope video recordings were paired with outcome data. Objects including durotomy, needle, grasper, needle driver, and nerve hook were then annotated. Altogether, SOSpine contains 15,698 frames with 53,238 annotations and associated durotomy repair outcomes. For validation, an AI model was fine-tuned on SOSpine video and detected surgical instruments with a mean average precision of 0.77. In summary, SOSpine depicts spine surgeons managing a common complication, providing opportunities to develop surgical AI models.

Silent Corticotroph Staining Pituitary Neuroendocrine Tumors: Prognostic Significance in Radiosurgery.

BACKGROUND AND OBJECTIVES: There is conflicting evidence on the significance of adrenocorticotrophic hormone (ACTH) staining in the prognosis of nonfunctioning pituitary neuroendocrine tumors (NFpitNETs). The objective of this study was to define the effect of ACTH immunostaining on clinical and radiographic outcomes of stereotactic radiosurgery (SRS) for NFpitNETs. METHODS: This retrospective, multicenter study included patients managed with SRS for NFpitNET residuals. The patients were divided into 2 cohorts: (1) silent corticotroph (SC) for NFpitNETs with positive ACTH immunostaining and (2) non-SC NFpitNETs. Rates of local tumor control and the incidence of post-treatment pituitary and neurological dysfunction were documented. Factors associated with radiological and clinical outcomes were also analyzed. RESULTS: The cohort included 535 patients from 14 centers with 84 (15.7%) patients harboring silent corticotroph NFpitNETs (SCs). At last follow-up, local tumor progression occurred in 11.9% of patients in the SC compared with 8.1% of patients in the non-SC cohort (P = .27). No statistically significant difference was noted in new-onset hypopituitarism rates (10.7% vs 15.4%, P = .25) or visual deficits (3.6% vs 1.1%, P = .088) between the 2 cohorts at last follow-up. When controlling for residual tumor volume, maximum dose, and patient age and sex, positive ACTH immunostaining did not have a significant correlation with local tumor progression (hazard ratio = 1.69, 95% CI = 0.8-3.61, P = .17). CONCLUSION: In contemporary radiosurgical practice with a single fraction dose of 8-25 Gy (median 15 Gy), ACTH immunostaining in NFpitNETs did not appear to confer a significantly reduced rate of local tumor control after SRS.

Pilot Analysis of Surgeon Instrument Utilization Signatures Based on Shannon Entropy and Deep Learning for Surgeon Performance Assessment in a Cadaveric Carotid Artery Injury Control Simulation.

BACKGROUND AND OBJECTIVES: Assessment and feedback are critical to surgical education, but direct observational feedback by experts is rarely provided because of time constraints and is typically only qualitative. Automated, video-based, quantitative feedback on surgical performance could address this gap, improving surgical training. The authors aim to demonstrate the ability of Shannon entropy (ShEn), an information theory metric that quantifies series diversity, to predict surgical performance using instrument detections generated through deep learning. METHODS: Annotated images from a publicly available video data set of surgeons managing endoscopic endonasal carotid artery lacerations in a perfused cadaveric simulator were collected. A deep learning model was implemented to detect surgical instruments across video frames. ShEn score for the instrument sequence was calculated from each surgical trial. Logistic regression using ShEn was used to predict hemorrhage control success. RESULTS: ShEn scores and instrument usage patterns differed between successful and unsuccessful trials (ShEn: 0.452 vs 0.370, P < .001). Unsuccessful hemorrhage control trials displayed lower entropy and less varied instrument use patterns. By contrast, successful trials demonstrated higher entropy with more diverse instrument usage and consistent progression in instrument utilization. A logistic regression model using ShEn scores (78% accuracy and 97% average precision) was at least as accurate as surgeons' attending/resident status and years of experience for predicting trial success and had similar accuracy as expert human observers. CONCLUSION: ShEn score offers a summative signal about surgeon performance and predicted success at controlling carotid hemorrhage in a simulated cadaveric setting. Future efforts to generalize ShEn to additional surgical scenarios can further validate this metric.

GPT-4 Artificial Intelligence Model Outperforms ChatGPT, Medical Students, and Neurosurgery Residents on Neurosurgery Written Board-Like Questions.

BACKGROUND: Artificial intelligence (AI) and machine learning have transformed health care with applications in various specialized fields. Neurosurgery can benefit from artificial intelligence in surgical planning, predicting patient outcomes, and analyzing neuroimaging data. GPT-4, an updated language model with additional training parameters, has exhibited exceptional performance on standardized exams. This study examines GPT-4's competence on neurosurgical board-style questions, comparing its performance with medical students and residents, to explore its potential in medical education and clinical decision-making. METHODS: GPT-4's performance was examined on 643 Congress of Neurological Surgeons Self-Assessment Neurosurgery Exam (SANS) board-style questions from various neurosurgery subspecialties. Of these, 477 were text-based and 166 contained images. GPT-4 refused to answer 52 questions that contained no text. The remaining 591 questions were inputted into GPT-4, and its performance was evaluated based on first-time responses. Raw scores were analyzed across subspecialties and question types, and then compared to previous findings on Chat Generative pre-trained transformer performance against SANS users, medical students, and neurosurgery residents. RESULTS: GPT-4 attempted 91.9% of Congress of Neurological Surgeons SANS questions and achieved 76.6% accuracy. The model's accuracy increased to 79.0% for text-only questions. GPT-4 outperformed Chat Generative pre-trained transformer (P < 0.001) and scored highest in pain/peripheral nerve (84%) and lowest in spine (73%) categories. It exceeded the performance of medical students (26.3%), neurosurgery residents (61.5%), and the national average of SANS users (69.3%) across all categories. CONCLUSIONS: GPT-4 significantly outperformed medical students, neurosurgery residents, and the national average of SANS users. The mode's accuracy suggests potential applications in educational settings and clinical decision-making, enhancing provider efficiency, and improving patient care.

Anatomic Considerations Guiding Single Versus Multiportal Endoscopic Approaches for Resection of Juvenile Nasopharyngeal Angiofibroma: Cases Series With Graded Multicorridor Resections.

BACKGROUND: Juvenile nasopharyngeal angiofibromas (JNAs) are characterized by expansive and destructive growth, often invading the midline/paranasal sinuses, pterygopalatine fossa, and infratemporal fossa and can extend into the orbit, cavernous sinus, or intracranially. OBJECTIVE: To evaluete the major benefits of the extended endoscopic endonasal approach (EEA) for JNA resection as compared with more traditional and invasive transpalatal and transfacial approaches. When JNAs extend into lateral anatomic compartments, the optimal operative trajectory often requires additional approach strategies or surgical staging. METHODS: We retrospectively reviewed 8 cases of large JNAs arising in symptomatic adolescent boys (University of Pittsburgh Medical Center Stages II, III, and V) and discuss anatomic and tumor considerations guiding the decision of a pure EEA vs combined EEA and sublabial transmaxillary approach (Caldwell-Luc). RESULTS: A pure extended EEA was used in 6 JNA cases (UPMC Stages II-III); a multiportal EEA + Caldwell-Luc maxillotomy was used in 2 cases. One of the 2 patients (UPMC Stage V) previously treated with multiportal EEA + Caldwell-Luc maxillotomy underwent staged left temporal/transzygomatic craniotomy, obtaining gross total resection. Seven patients ultimately underwent complete removal without recurrence. One patient with a small residual JNA (UPMC II) underwent stereotactic radiosurgery without progression to date. CONCLUSION: JNAs with lateral extension into the infratemporal fossa often benefited from additional lateral exposure using a Caldwell-Luc maxillotomy. Cases with significant skull base and/or dural involvement may undergo staged surgical treatment; temporalis + transzygomatic craniotomy is often useful for second-stage approaches for residual tumor in these lateral infratemporal or intracranial regions. SRS should be considered for residual tumor if additional surgery is not warranted.

Validation of Myc-Associated Protein X (MAX) regulation in growth hormone secreting and nonfunctional pituitary adenoma.

INTRODUCTION: Many patients with growth hormone-secreting pituitary adenoma (GHPA) fail to achieve biochemical remission, warranting investigation into epigenetic and molecular signatures associated with tumorigenesis and hormonal secretion. Prior work exploring the DNA methylome showed Myc-Associated Protein X (MAX), a transcription factor involved in cell cycle regulation, was differentially methylated between GHPA and nonfunctional pituitary adenoma (NFPA). We aimed to validate the differential DNA methylation and related MAX protein expression profiles between NFPA and GHPA. METHODS: DNA methylation levels were measured in 52 surgically resected tumors (37 NFPA, 15 GHPA) at ~100,000 known MAX binding sites derived using ChIP-seq analysis from ENCODE. Findings were correlated with MAX protein expression using a constructed tissue microarray (TMA). Gene ontology analysis was performed to explore downstream genetic and signaling pathways regulated by MAX. RESULTS: GHPA had more hypomethylation events across all known MAX binding sites. Of binding sites defined using ChIP-seq analysis, 1,551 sites had significantly different methylation patterns between the two cohorts; 432 occurred near promoter regions potentially regulated by MAX, including promoters of TNF and MMP9. Gene ontology analysis suggested enrichment in genes involved in oxygen response, immune system regulation, and cell proliferation. Thirteen MAX binding sites were within coding regions of genes. GHPA demonstrated significantly increased expression of MAX protein compared to NFPA. CONCLUSION: GHPA have significantly different DNA methylation and downstream protein expression levels of MAX compared to NFPA. These differences may influence mechanisms involved with cellular proliferation, tumor invasion and hormonal secretion.

From white to blue light: evolution of endoscope-assisted intracranial tumor neurosurgery and expansion to intraaxial tumors.

OBJECTIVE: Intraoperative use of the endoscope to assist in visualization of intracranial tumor pathology has expanded with increasing surgeon experience and improved instrumentation. The authors aimed to study how advancements in endoscopic technology have affected the evolution of endoscope use, with particular focus on blue light-filter modification allowing for discrimination of fluorescent tumor tissue following 5-ALA administration. METHODS: A retrospective analysis of patients undergoing craniotomy for tumor resection at a single institution between February 2012 and July 2021 was performed. Patients were included if the endoscope was used for diagnostic tumor cavity inspection or therapeutic assistance with tumor resection following standard craniotomy and microsurgical tumor resection, with emphasis on those cases in which blue light endoscopy was used. Medical records were queried for patient demographics, operative reports describing the use of the endoscope and extent of resection, associations with tumor pathology, and postoperative outcomes. Preoperative and postoperative MR images were reviewed for radiographic extent of resection. RESULTS: A total of 52 patients who underwent endoscope-assisted craniotomy for tumor were included. Thirty patients (57.7%) were men and the average age was 52.6 ± 16.1 years. Standard white light endoscopes were used for assistance with tumor resection in 28 cases (53.8%) for tumors primarily located in the ventricular system, parasellar region, and cerebellopontine angle. A blue light endoscope for detection of 5-ALA fluorescence was introduced into our practice in 2014 and subsequently used for assistance with tumor resection in 24 cases (46.2%) (intraaxial: n = 22, extraaxial: n = 2). Beyond the use of the surgical microscope as the primary visualization source, the blue light endoscope was used to directly perform additional tumor resection in 19/21 cases as a result of improved fluorescence detection as compared to the surgical microscope. No complications were associated with the use of the endoscope or with additional resection performed under white or blue light visualization. CONCLUSIONS: Endoscopic assistance to visualize intracranial tumors had previously been limited to white light, assisting mostly in the visualization of extraaxial tumors confined to intraventricular and cisternal compartments. Blue light-equipped endoscopes provide improved versatility and visualization of 5-ALA fluorescing tissue beyond the capability of the surgical microscope, thereby expanding its use into the realm of intraaxial tumor resections.

Adjuvant versus on-progression Gamma Knife radiosurgery for residual nonfunctioning pituitary adenomas: a matched-cohort analysis.

OBJECTIVE: Radiological progression occurs in 50%-60% of residual nonfunctioning pituitary adenomas (NFPAs). Stereotactic radiosurgery (SRS) is a safe and effective management option for residual NFPAs, but there is no consensus on its optimal timing. This study aims to define the optimal timing of SRS for residual NFPAs. METHODS: This retrospective, multicenter study involved 375 patients with residual NFPAs managed with SRS. The patients were divided into adjuvant (ADJ; treated for stable residual NFPA within 6 months of resection) and progression (PRG) cohorts (treated for residual NFPA progression). Factors associated with tumor progression and clinical deterioration were analyzed. RESULTS: Following propensity-score matching, each cohort consisted of 130 patients. At last follow-up, tumor control was achieved in 93.1% of patients in the ADJ cohort and in 96.2% of patients in the PRG cohort (HR 1.6, 95% CI 0.55-4.9, p = 0.37). Hypopituitarism was associated with a maximum point dose of > 8 Gy to the pituitary stalk (HR 4.5, 95% CI 1.6-12.6, p = 0.004). No statistically significant difference was noted in crude new-onset hypopituitarism rates (risk difference [RD] = -0.8%, p > 0.99) or visual deficits (RD = -2.3%, p = 0.21) between the two cohorts at the last follow-up. The median time from resection to new hypopituitarism was longer in the PRG cohort (58.9 vs 29.7 months, p = 0.01). CONCLUSIONS: SRS at residual NFPA progression does not appear to alter the probability of tumor control or hormonal/visual deficits compared with adjuvant SRS. Deferral of radiosurgical management to the time of radiological progression could significantly prolong the time to radiosurgically induced pituitary dysfunction. A lower maximum point dose (< 8 Gy) to the pituitary stalk portended a more favorable chance of preserving pituitary function after SRS.

A quantitative characterization of the spatial distribution of brain metastases from breast cancer and respective molecular subtypes.

PURPOSE: Brain metastases (BM) remain a significant cause of morbidity and mortality in breast cancer (BC) patients. Specific factors promoting the process of BM and predilection for selected neuro-anatomical regions remain unknown, yet may have major implications for prevention or treatment. Anatomical spatial distributions of BM from BC suggest a predominance of metastases in the hindbrain and cerebellum. Systematic approaches to quantifying BM location or location-based analyses based on molecular subtypes, however, remain largely unavailable. METHODS: We analyzed stereotactic Cartesian coordinates derived from 134 patients undergoing gamma- knife radiosurgery (GKRS) for treatment of 407 breast cancer BMs to quantitatively study BM spatial distribution along principal component axes and by intrinsic molecular subtype (ER, PR, Herceptin). We used kernel density estimators (KDE) to highlight clustering and distribution regions in the brain, and we used the metric of mutual information (MI) to tease out subtle differences in the BM distributions associated with different molecular subtypes of BC. BM location maps according to vascular and anatomical distributions using Cartesian coordinates to aid in systematic classification of tumor locations were additionally developed. RESULTS: We corroborated that BC BMs show a consistent propensity to arise posteriorly and caudally, and that Her2+ tumors are relatively more likely to arise medially rather than laterally. To compare the distributions among varying BC molecular subtypes, the mutual information metric reveal that the ER-PR-Her2+ and ER-PR-Her2- subtypes show the smallest amount of mutual information and are most molecularly distinct. The kernel density contour plots show a propensity for triple negative BC to arise in more superiorly or cranially situated BMs. CONCLUSIONS: We present a novel and shareable workflow for characterizing and comparing spatial distributions of BM which may aid in identifying therapeutic or diagnostic targets and interactions with the tumor microenvironment. Further characterization of these patterns with larger multi-institutional data-sets may have major impacts on treatment or management of cancer patients.

The abscopal effect: systematic review in patients with brain and spine metastases.

Background: The abscopal effect is a rare phenomenon whereby local radiation induces a proposed immune-mediated anti-tumor effect at distant sites. Given the growing use of immunotherapies and systemic immune checkpoint inhibitors in neuro-oncologic practice, we aimed to review prior studies pertaining to this phenomenon in the context of tumor shrinkage both within the central nervous system as well as distant disease sites. Methods: A systematic review in accordance with the PRISMA guidelines was conducted to identify all studies which assessed the abscopal effect in patients with treated metastatic cancer to the brain and/or spine. Articles were included if they reported the abscopal effect in patients (case studies) or if the abscopal effect was explicitly analyzed in case series with cohorts of patients with metastatic brain or spine tumors. Laboratory investigations and clinical trials investigating new therapies were excluded. Results: Twenty reports met inclusion criteria [16 case reports, 4 case series (n = 160), total n = 174]. Case reports of the abscopal effect were in relation to the following cancers: melanoma (6 patients), breast cancer (3), lung adenocarcinoma (2), non-small-cell lung cancer (2), hepatocellular carcinoma (1), and renal cell carcinoma (1). Eleven patients had irradiation to the brain and 2 to the spine. Patients undergoing whole brain radiotherapy (6) had an average dose of 33.6 Gy over 8-15 fractions, and those undergoing stereotactic radiosurgery (5) had an average dose of 21.5 Gy over 1-5 fractions. One patient had radiation to the body and an intracranial abscopal effect was observed. Most common sites of extracranial tumor reduction were lung and lymph nodes. Ten case studies (57%) showed complete resolution of extra-CNS tumor burden. Median progression-free survival was 13 months following radiation. Four papers investigated incidence of abscopal effects in patients with metastatic melanoma to the brain who received immune checkpoint inhibitor therapy (n = 160); two papers found an abscopal effect in 35% and 52% of patients (n = 16, 21 respectively), and two papers found no evidence of abscopal effects (n = 61, 62). Conclusions: Abscopal effects can occur following radiotherapy in patients with brain or spine metastases and is thought to be a result of increased anti-tumor immunity. The potential for immune checkpoint inhibitor therapy to be used in combination with radiotherapy to induce an abscopal effect is an area of active investigation.

Validation of Machine Learning-Based Automated Surgical Instrument Annotation Using Publicly Available Intraoperative Video.

BACKGROUND: Intraoperative tool movement data have been demonstrated to be clinically useful in quantifying surgical performance. However, collecting this information from intraoperative video requires laborious hand annotation. The ability to automatically annotate tools in surgical video would advance surgical data science by eliminating a time-intensive step in research. OBJECTIVE: To identify whether machine learning (ML) can automatically identify surgical instruments contained within neurosurgical video. METHODS: A ML model which automatically identifies surgical instruments in frame was developed and trained on multiple publicly available surgical video data sets with instrument location annotations. A total of 39 693 frames from 4 data sets were used (endoscopic endonasal surgery [EEA] [30 015 frames], cataract surgery [4670], laparoscopic cholecystectomy [2532], and microscope-assisted brain/spine tumor removal [2476]). A second model trained only on EEA video was also developed. Intraoperative EEA videos from YouTube were used for test data (3 videos, 1239 frames). RESULTS: The YouTube data set contained 2169 total instruments. Mean average precision (mAP) for instrument detection on the YouTube data set was 0.74. The mAP for each individual video was 0.65, 0.74, and 0.89. The second model trained only on EEA video also had an overall mAP of 0.74 (0.62, 0.84, and 0.88 for individual videos). Development costs were $130 for manual video annotation and under $100 for computation. CONCLUSION: Surgical instruments contained within endoscopic endonasal intraoperative video can be detected using a fully automated ML model. The addition of disparate surgical data sets did not improve model performance, although these data sets may improve generalizability of the model in other use cases.

Expert surgeons and deep learning models can predict the outcome of surgical hemorrhage from 1 min of video.

Major vascular injury resulting in uncontrolled bleeding is a catastrophic and often fatal complication of minimally invasive surgery. At the outset of these events, surgeons do not know how much blood will be lost or whether they will successfully control the hemorrhage (achieve hemostasis). We evaluate the ability of a deep learning neural network (DNN) to predict hemostasis control ability using the first minute of surgical video and compare model performance with human experts viewing the same video. The publicly available SOCAL dataset contains 147 videos of attending and resident surgeons managing hemorrhage in a validated, high-fidelity cadaveric simulator. Videos are labeled with outcome and blood loss (mL). The first minute of 20 videos was shown to four, blinded, fellowship trained skull-base neurosurgery instructors, and to SOCALNet (a DNN trained on SOCAL videos). SOCALNet architecture included a convolutional network (ResNet) identifying spatial features and a recurrent network identifying temporal features (LSTM). Experts independently assessed surgeon skill, predicted outcome and blood loss (mL). Outcome and blood loss predictions were compared with SOCALNet. Expert inter-rater reliability was 0.95. Experts correctly predicted 14/20 trials (Sensitivity: 82%, Specificity: 55%, Positive Predictive Value (PPV): 69%, Negative Predictive Value (NPV): 71%). SOCALNet correctly predicted 17/20 trials (Sensitivity 100%, Specificity 66%, PPV 79%, NPV 100%) and correctly identified all successful attempts. Expert predictions of the highest and lowest skill surgeons and expert predictions reported with maximum confidence were more accurate. Experts systematically underestimated blood loss (mean error - 131 mL, RMSE 350 mL, R2 0.70) and fewer than half of expert predictions identified blood loss > 500 mL (47.5%, 19/40). SOCALNet had superior performance (mean error - 57 mL, RMSE 295 mL, R2 0.74) and detected most episodes of blood loss > 500 mL (80%, 8/10). In validation experiments, SOCALNet evaluation of a critical on-screen surgical maneuver and high/low-skill composite videos were concordant with expert evaluation. Using only the first minute of video, experts and SOCALNet can predict outcome and blood loss during surgical hemorrhage. Experts systematically underestimated blood loss, and SOCALNet had no false negatives. DNNs can provide accurate, meaningful assessments of surgical video. We call for the creation of datasets of surgical adverse events for quality improvement research.

Association between socioeconomic status and presenting characteristics and extent of disease in patients with surgically resected nonfunctioning pituitary adenoma.

OBJECTIVE: The aim of this study was to evaluate the association between zip code-level socioeconomic status (SES) and presenting characteristics and short-term clinical outcomes in patients with nonfunctioning pituitary adenoma (NFPA). METHODS: A retrospective review of prospectively collected data from the University of Southern California Pituitary Center was conducted to identify all patients undergoing surgery for pituitary adenoma (PA) from 2000 to 2021 and included all patients with NFPA with recorded zip codes at the time of surgery. A normalized socioeconomic metric by zip code was then constructed using data from the American Community Survey estimates, which was categorized into tertiles. Multiple imputation was used for missing data, and multivariable linear and logistic regression models were constructed to estimate mean differences and multivariable-adjusted odds ratios for the association between zip code-level SES and presenting characteristics and outcomes. RESULTS: A total of 637 patients were included in the overall analysis. Compared with patients in the lowest SES tertile, those in the highest tertile were more likely to be treated at a private (rather than safety net) hospital, and were less likely to present with headache, vision loss, and apoplexy. After multivariable adjustment for age, sex, and prior surgery, SES in the highest compared with lowest tertile was inversely associated with tumor size at diagnosis (-4.9 mm, 95% CI -7.2 to -2.6 mm, p < 0.001) and was positively associated with incidental diagnosis (multivariable-adjusted OR 1.72, 95% CI 1.02-2.91). Adjustment for hospital (private vs safety net) attenuated the observed associations, but disparities by SES remained statistically significant for tumor size. Despite substantial differences at presentation, there were no significant differences in length of stay or odds of an uncomplicated procedure by zip code-level SES. Patients from lower-SES zip codes were more likely to require postoperative steroid replacement and less likely to achieve gross-total resection. CONCLUSIONS: In this series, lower zip code-level SES was associated with more severe disease at the time of diagnosis for NFPA patients, including larger tumor size and lower rates of incidental diagnosis. Despite these differences at presentation, no significant differences were observed in short-term postoperative complications, although patients with higher zip code-level SES had higher rates of gross-total resection.

Code-free machine learning for object detection in surgical video: a benchmarking, feasibility, and cost study.

OBJECTIVE: While the utilization of machine learning (ML) for data analysis typically requires significant technical expertise, novel platforms can deploy ML methods without requiring the user to have any coding experience (termed AutoML). The potential for these methods to be applied to neurosurgical video and surgical data science is unknown. METHODS: AutoML, a code-free ML (CFML) system, was used to identify surgical instruments contained within each frame of endoscopic, endonasal intraoperative video obtained from a previously validated internal carotid injury training exercise performed on a high-fidelity cadaver model. Instrument-detection performances using CFML were compared with two state-of-the-art ML models built using the Python coding language on the same intraoperative video data set. RESULTS: The CFML system successfully ingested surgical video without the use of any code. A total of 31,443 images were used to develop this model; 27,223 images were uploaded for training, 2292 images for validation, and 1928 images for testing. The mean average precision on the test set across all instruments was 0.708. The CFML model outperformed two standard object detection networks, RetinaNet and YOLOv3, which had mean average precisions of 0.669 and 0.527, respectively, in analyzing the same data set. Significant advantages to the CFML system included ease of use, relatively low cost, displays of true/false positives and negatives in a user-friendly interface, and the ability to deploy models for further analysis with ease. Significant drawbacks of the CFML model included an inability to view the structure of the trained model, an inability to update the ML model once trained with new examples, and the inability for robust downstream analysis of model performance and error modes. CONCLUSIONS: This first report describes the baseline performance of CFML in an object detection task using a publicly available surgical video data set as a test bed. Compared with standard, code-based object detection networks, CFML exceeded performance standards. This finding is encouraging for surgeon-scientists seeking to perform object detection tasks to answer clinical questions, perform quality improvement, and develop novel research ideas. The limited interpretability and customization of CFML models remain ongoing challenges. With the further development of code-free platforms, CFML will become increasingly important across biomedical research. Using CFML, surgeons without significant coding experience can perform exploratory ML analyses rapidly and efficiently.

Deep Neural Networks Can Accurately Detect Blood Loss and Hemorrhage Control Task Success From Video.

BACKGROUND: Deep neural networks (DNNs) have not been proven to detect blood loss (BL) or predict surgeon performance from video. OBJECTIVE: To train a DNN using video from cadaveric training exercises of surgeons controlling simulated internal carotid hemorrhage to predict clinically relevant outcomes. METHODS: Video was input as a series of images; deep learning networks were developed, which predicted BL and task success from images alone (automated model) and images plus human-labeled instrument annotations (semiautomated model). These models were compared against 2 reference models, which used average BL across all trials as its prediction (control 1) and a linear regression with time to hemostasis (a metric with known association with BL) as input (control 2). The root-mean-square error (RMSE) and correlation coefficients were used to compare the models; lower RMSE indicates superior performance. RESULTS: One hundred forty-three trials were used (123 for training and 20 for testing). Deep learning models outperformed controls (control 1: RMSE 489 mL, control 2: RMSE 431 mL, R2 = 0.35) at BL prediction. The automated model predicted BL with an RMSE of 358 mL (R2 = 0.4) and correctly classified outcome in 85% of trials. The RMSE and classification performance of the semiautomated model improved to 260 mL and 90%, respectively. CONCLUSION: BL and task outcome classification are important components of an automated assessment of surgical performance. DNNs can predict BL and outcome of hemorrhage control from video alone; their performance is improved with surgical instrument presence data. The generalizability of DNNs trained on hemorrhage control tasks should be investigated.

Utility of the Simulated Outcomes Following Carotid Artery Laceration Video Data Set for Machine Learning Applications.

Importance: Surgical data scientists lack video data sets that depict adverse events, which may affect model generalizability and introduce bias. Hemorrhage may be particularly challenging for computer vision-based models because blood obscures the scene. Objective: To assess the utility of the Simulated Outcomes Following Carotid Artery Laceration (SOCAL)-a publicly available surgical video data set of hemorrhage complication management with instrument annotations and task outcomes-to provide benchmarks for surgical data science techniques, including computer vision instrument detection, instrument use metrics and outcome associations, and validation of a SOCAL-trained neural network using real operative video. Design, Setting, and Participants: For this quailty improvement study, a total of 75 surgeons with 1 to 30 years' experience (mean, 7 years) were filmed from January 1, 2017, to December 31, 2020, managing catastrophic surgical hemorrhage in a high-fidelity cadaveric training exercise at nationwide training courses. Videos were annotated from January 1 to June 30, 2021. Interventions: Surgeons received expert coaching between 2 trials. Main Outcomes and Measures: Hemostasis within 5 minutes (task success, dichotomous), time to hemostasis (in seconds), and blood loss (in milliliters) were recorded. Deep neural networks (DNNs) were trained to detect surgical instruments in view. Model performance was measured using mean average precision (mAP), sensitivity, and positive predictive value. Results: SOCAL contains 31 443 frames with 65 071 surgical instrument annotations from 147 trials with associated surgeon demographic characteristics, time to hemostasis, and recorded blood loss for each trial. Computer vision-based instrument detection methods using DNNs trained on SOCAL achieved a mAP of 0.67 overall and 0.91 for the most common surgical instrument (suction). Hemorrhage control challenges standard object detectors: detection of some surgical instruments remained poor (mAP, 0.25). On real intraoperative video, the model achieved a sensitivity of 0.77 and a positive predictive value of 0.96. Instrument use metrics derived from the SOCAL video were significantly associated with performance (blood loss). Conclusions and Relevance: Hemorrhage control is a high-stakes adverse event that poses unique challenges for video analysis, but no data sets of hemorrhage control exist. The use of SOCAL, the first data set to depict hemorrhage control, allows the benchmarking of data science applications, including object detection, performance metric development, and identification of metrics associated with outcomes. In the future, SOCAL may be used to build and validate surgical data science models.