In a recent review in Nature Medicine, researchers investigated using artificial intelligence (AI) in surgery, highlighting improvements in preoperative, intraoperative, and postoperative care.
Study: Artificial intelligence in surgery. Image Credit: LALAKA/Shutterstock.com
Background
AI is making significant advances in healthcare, including foundation model architectures, wearable technologies, and surgical data infrastructures. Research shows that artificial intelligence can exceed or complement human skills, notably in radiology.
However, surgery is still a slow-growing specialty, with global differences in access, complications, and post-complication mortality.
A comprehensive approach to surgical system strengthening is required, including enhanced access, education, issue detection and treatment, and system efficiency improvement.
About the review
In the present review, researchers emphasized the promise of AI technologies to improve patient outcomes, surgical education, and care optimization, emphasizing existing deep learning applications and anticipating future developments using multimodal foundation models.
Preoperative surgical applications of AI
Artificial intelligence (AI) is utilized in surgical planning and patient selection, notably in preoperative imaging, to aid in early diagnostic evaluations and planning. Model-free methods for reinforcement learning have shown potential for detecting and removing tumor tissue while limiting the effects on functioning biological tissues during neurosurgery.
Innovative techniques for preoperative planning use AI- and virtual reality-based segmentation algorithms, significantly improving surgical techniques.
A precise preoperative clinical diagnosis is essential for making decisions and planning treatment. The RadioLOGIC AI algorithm pulls unstructured data from radiological reports in medical health records to improve radiological diagnosis.
Progress with large language models (LLMs) and interaction with electronic data systems may allow for earlier illness detection and treatment before the disease progresses.
Diagnosis is a highly developed aspect of surgical artificial intelligence, with model generalizability and accuracy finding early clinical use. Task-specific breakthroughs provide more precise tumor staging, potentially improving surgical planning.
Intraoperative surgical applications of AI
In surgical practice, the intraoperative phase is a data-rich environment that continuously monitoring physiological indicators and complicated insults.
Intraoperative computer vision advancements have allowed for advancements in anatomy analysis, tissue feature evaluation, dissection plane assessment, pathology detection, and trustworthy instrument identification. However, operating theaters record, evaluate, and gather limited data.
Researchers must use valuable intraoperative phase data streams to help with surgical automation and artificial intelligence in the operating theater.
AI in surgical decision-making helps enhance surgical resection margins, shorten operating times, and increase efficiency. A recent patient-agnostic transfer-learned neural network uses quick nanopore sequencing to provide accurate intraoperative diagnoses in under 40 minutes.
Multimodal AI interrogation might help determine pertinent anatomy, augment surgeons’ visual assessments, inform biopsies, and quantify cancer risk.
A digitized surgical platform may pave the way for an AI-enhanced future, with platform investments critical to capitalizing on digital advancements. Intraoperative apps are vital to the times to come for surgery since they improve nontechnical activities like communication, cooperation, and competence assessment.
Postoperative surgical applications of AI
Hospital-at-home care aims to enhance healthcare access and equity by allowing patients to recuperate in familiar settings and resume normal functioning within society.
While there has been progress in decreasing postoperative duration of stay, enabling early discharge, and facilitating functional recovery using minimally invasive-type surgical methods, early return to routine activities, improved postoperative evaluation, and early warning provision systems, data-based innovations are lacking in the postoperative period.
Wearables provide continuous monitoring by allowing for multimodal physiological indicator inputs, which can help with data-driven discharge planning. A systematic evaluation identified 31 wearable devices that monitor physiological data, vital signs, and physical activity.
Complication prediction following surgery is difficult due to the numerous variables influencing care and outcomes. Early diagnosis of complications, particularly life-threatening ones such as anastomotic leaks and pancreatic fistulas, is critical for healthcare systems to reduce mortality.
MySurgeryRisk is one of the few advancements in complication prediction using machine learning algorithms; however, there is limited knowledge of the scalability of these algorithms to other health systems.
Researchers developed an AI-driven home-based rehabilitation model that incorporates real-time data gathering, sophisticated evaluation of daily living measures, and novel assessments of ADLs.
Conclusion
Based on the review findings, AI technology can improve surgical treatment by optimizing patient selection, intraoperative performance, and operating room procedures. Transformers, a neural network design breakthrough, have allowed multimodal AI models with significant surgical applications.
These include clinical risk prediction, automation, computer vision in robotic surgery, intraoperative diagnostics, enhanced training, sensor-based postoperative monitoring, resource management, and discharge planning.
However, thorough scrutiny and regulatory oversight are required, and stakeholder participation is critical to providing improved surgical care. Research must guide AI systems by providing benefits such as accurate diagnosis and improved system efficiency.
This article was originally published on news-medical