A new dataset and versatile multi-task surgical workflow analysis framework for thoracoscopic mitral valvuloplasty.

Few-shot learning for surgical phase recognition: Performance and generalization in cholecystectomy.

Surgical Phase Recognition (SPR) enables real-time, context-aware assistance during surgery, but its use remains limited by the cost and effort of dense video annotation. This study presents a Semi-Supervised Deep Learning framework for SPR in endoscopic pituitary surgery, aiming to reduce annotation requirements while maintaining performance. A Timestamp Supervision strategy is employed, where only one or two representative frames per phase are labeled. These labels are then propagated, creating pseudo-labels for unlabeled frames using an Uncertainty-Aware Temporal Diffusion (UATD) approach, based on confidence and temporal consistency. Multiple spatial and temporal architectures are evaluated on the PituPhase–SurgeryAI dataset, the largest publicly available collection of endoscopic pituitary surgeries to date, which includes an outside-the-body phase. Despite using less than 3% of the annotated data, the proposed method achieves an F1-score of 0.60 [0.55–0.65], demonstrating competitive performance against previous Supervised approaches in the same context. Removing the recurrent outside-the-body phase reduces misclassification and improves temporal consistency. These results demonstrate that uncertainty-guided Semi-Supervision is a scalable and clinically viable alternative to fully Supervised Learning for surgical workflow analysis.

Various complications of pituitary neuroendocrine tumors (PitNET) are reported, and an intratumor hemorrhage or infarct underlying pituitary apoplexy (PA) represents an uncommon, yet potentially life-threatening, feature, and thus early recognition and prompt intervention are important. Our purpose is to overview PA from clinical presentation to management and outcome. This is a narrative review of the English-language, PubMed-based original articles from 2012 to 2022 concerning PA, with the exception of pregnancy- and COVID-19-associated PA, and non-spontaneous PA (prior specific therapy for PitNET). We identified 194 original papers including 1452 patients with PA (926 males, 525 females, and one transgender male; a male-to-female ratio of 1.76; mean age at PA diagnostic of 50.52 years, the youngest being 9, the oldest being 85). Clinical presentation included severe headache in the majority of cases (but some exceptions are registered, as well); neuro-ophthalmic panel with nausea and vomiting, meningism, and cerebral ischemia; respectively, decreased visual acuity to complete blindness in two cases; visual field defects: hemianopia, cranial nerve palsies manifesting as diplopia in the majority, followed by ptosis and ophthalmoplegia (most frequent cranial nerve affected was the oculomotor nerve, and, rarely, abducens and trochlear); proptosis (N = 2 cases). Risk factors are high blood pressure followed by diabetes mellitus as the main elements. Qualitative analysis also pointed out infections, trauma, hematologic conditions (thrombocytopenia, polycythemia), Takotsubo cardiomyopathy, and T3 thyrotoxicosis. Iatrogenic elements may be classified into three main categories: medication, diagnostic tests and techniques, and surgical procedures. The first group is dominated by anticoagulant and antiplatelet drugs; additionally, at a low level of statistical evidence, we mention androgen deprivation therapy for prostate cancer, chemotherapy, thyroxine therapy, oral contraceptives, and phosphodiesterase 5 inhibitors. The second category includes a dexamethasone suppression test, clomiphene use, combined endocrine stimulation tests, and a regadenoson myocardial perfusion scan. The third category involves major surgery, laparoscopic surgery, coronary artery bypass surgery, mitral valvuloplasty, endonasal surgery, and lumbar fusion surgery in a prone position. PA in PitNETs still represents a challenging condition requiring a multidisciplinary team from first presentation to short- and long-term management. Controversies involve the specific panel of risk factors and adequate protocols with concern to neurosurgical decisions and their timing versus conservative approach. The present decade-based analysis, to our knowledge the largest so far on published cases, confirms a lack of unanimous approach and criteria of intervention, a large panel of circumstantial events, and potential triggers with different levels of statistical significance, in addition to a heterogeneous clinical picture (if any, as seen in subacute PA) and a spectrum of evolution that varies from spontaneous remission and control of PitNET-associated hormonal excess to exitus. Awareness is mandatory. A total of 25 cohorts have been published so far with more than 10 PA cases/studies, whereas the largest cohorts enrolled around 100 patients. Further studies are necessary.

Surgical videos are a valuable data source, offering detailed insights into surgical practice. However, video analysis requires specialist clinical knowledge and takes considerable time. Artificial intelligence (AI) has the potential to improve and streamline the interpretation of intraoperative video data. This systematic scoping review aimed to summarize the use of AI in the analysis of videos of surgical procedures and identify evidence gaps. Systematic searches of Ovid MEDLINE and Embase were performed using search terms 'artificial intelligence', 'video', and 'surgery'. Data extraction included reporting of general study characteristics; the overall objective of AI; descriptions of data sets, AI models, and training; methods of data annotation; and measures of accuracy. Data were summarized descriptively. In all, 122 studies were included. More than half focused on gastrointestinal procedures (75 studies, 61.5%), predominantly cholecystectomy (47, 38.5%). The most common objectives were surgical phase recognition (40 studies, 32.8%), surgical instrument recognition (28, 23.0%), and enhanced intraoperative visualization (23, 18.9%). Of the studies, 79.5% (97) used a single data set and most (92, 75.4%) used supervised machine learning techniques. There was considerable variation across the studies in terms of the number of videos, centres, and contributing surgeons. Forty-seven studies (38.5%) did not report the number of annotators, and details about their experience were frequently omitted (102, 83.6%). Most studies used multiple outcome measures (67, 54.9%), most commonly overall or best accuracy of the AI model (67, 54.9%). This review found that many studies omitted essential methodological details of AI training, testing, data annotation, and validation processes, creating difficulties when interpreting and replicating these studies. Another key finding was the lack of large data sets from multiple centres and surgeons. Future research should focus on curating large, varied, open-access data sets from multiple centres, patients, and surgeons to facilitate accurate evaluation using real-world data.

Deep reinforcement learning has achieved significant success in solving sequential decision-making tasks. Excellent models usually require the input of valid state signals during training, which is challenging to encode temporal state features for the deep reinforcement learning model. To address this issue, recent methods attempt to encode multi-step sequential state signals so as to obtain more comprehensive observational information. However, these methods usually have a lower performance on complex continuous control tasks because mapping the state sequence into a low-dimensional embedding causes blurring of the immediate state features. In this paper, we propose a multiple frequency bands temporal state representation learning framework. The temporal state signals are decomposed into discrete state signals of various frequency bands by Discrete Fourier Transform (DFT). Then, feature signals filtered out different high-frequency bands are generated. Meanwhile, the mask generator evaluates the weights of signals of various frequency bands and encodes high-quality representations for agent training. Our intuition is that temporal state representations considering multiple frequency bands have high fidelity and stability. We conduct experiments tasks and verify that our method has obvious advantages over the baseline in complex continuous control tasks such as Walker and Crawler.

Intraoperative video analysis and machine learning models will change the future of surgical training

Manipulating the expression of synaptic plasticity in neuromorphic devices provides essential foundations for developing intelligent, adaptive hardware systems. In recent years, advances have shifted from static emulation toward dynamic, network-oriented plasticity design, offering enhanced computational accuracy and functional relevance. This review highlights how diversified plasticity behaviors, including multilevel long-term potentiation and depression for spatial models, tunable short-term memory for temporal models, as well as wavelength-selective response, excitatory and inhibitory synergy, and adaptive threshold modulation, collectively support key tasks such as stable learning, temporal processing, and context-aware adaptation. Beyond behavioral innovations, strategies such as multifunctional single-device integration, multimodal fusion, and heterogeneous system assembly enable compact, energy-efficient, and versatile neuromorphic architectures. Recent developments at the array level further demonstrate high-performance scalability and system-level applicability. Despite notable progress, current modulation strategies remain constrained in flexibility, diversity, and large-scale coordination. Future research should focus on enriching the behavioral repertoire of plasticity, advancing cross-modal convergence, and improving array-level uniformity, paving the way toward deployable, high-efficiency neuromorphic intelligence.

Artificial intelligence (AI) technology has been used to assess surgery quality, educate, and evaluate surgical performance using video recordings in the minimally invasive surgery era. Much attention has been paid to automating surgical workflow analysis from surgical videos for an effective evaluation to achieve the assessment and evaluation. This study aimed to design a deep learning model to automatically identify surgical phases using laparoscopic cholecystectomy videos and automatically assess the accuracy of recognizing surgical phases. One hundred and twenty cholecystectomy videos from a public dataset (Cholec80) and 40 laparoscopic cholecystectomy videos recorded between July 2022 and December 2022 at a single institution were collected. These datasets were split into training and testing datasets for the AI model at a 2:1 ratio. Test scenarios were constructed according to structural characteristics of the trained model. No pre- or post-processing of input data or inference output was performed to accurately analyze the effect of the label on model training. A total of 98,234 frames were extracted from 40 cases as test data. The overall accuracy of the model was 91.2%. The most accurate phase was Calot's triangle dissection (F1 score: 0.9421), whereas the least accurate phase was clipping and cutting (F1 score: 0.7761). Our AI model identified phases of laparoscopic cholecystectomy with a high accuracy.

Surgical tool recognition is a key task to analyze surgical workflow, in order to improve the efficiency and safety of laparoscopic surgeries. The laparoscopic videos are important sources to conduct this task, However, there are some challenges to analyze these videos. Focus on the imbalanced dataset problem, data augmentation method based on generate different synthetic datasets and evaluate their performance training on a convolutional neural network model are investigated in this research. The results show the effect on the model with different background patterns. A better performance was achieved when the model was trained by a structure background dataset. Further research will be needed to understand why the original background patterns support the correct classification. It is assumed that this is an overlearning effect, that will not hold if other procedures were included into the test set.