Artificial Intelligence and Interventional Surgical Robots

Healthcare Author: Yan Zhang Editor: Mianmian Wang Nov 03, 2022 06:11 PM (GMT+8)

The evolvement and implementation of artificial intelligence (AI) are dramatically changing the medical landscape, and the potential benefits of using AI in interventional medicine are now also being extensively pursued. The article focuses primarily on advances and challenges in AI applied to interventional surgical robots.

artificial intelligence

What does AI bring to interventional surgical robots?

Interventional surgical robots remove the physician from X-ray hazards, enable surgeries and stenting without compromising safety, and allow increased precision. Image navigation is the eye and brain of interventional robots, playing a crucial role in both diagnoses and as the primary guidance tool during interventions. Fortunately, powerful artificial intelligence (AI) technology is penetrating the medical imaging arena, holding significant promise for creating an 'eye-hand-brain' collaborative system for interventional robots and optimizing fluoroscopic interventional procedures.

From preoperative treatment plans to intraoperative imaging navigation and postoperative imaging follow-ups, AI can help realize image-guided precision medical visualization and provide physicians with additional information not available through conventional approaches.

Pre, during and post-intervention.jpg.jpg

Fluoroscopy and angiography are primary imaging modalities during percutaneous coronary interventions (PCI). Routine tasks based on medical imaging include anatomical classification, detection, segmentation, and registration.

With pre-intervention 3D medical data, physicians can reconstruct and analyze multimodality 3D imaging of vital organs and tumors, after which a preoperative treatment plan can be set in the corresponding 3D interactive interface. Computer simulation methods enable pre-intervention simulation of the clinical execution of operation plans and provide physicians with reliable surgical guidance. Any form of comprehensive utilization of multimodal image information is called fusion. Fusion of the angiographic images with CT data enhances the 3D performance of procedures such as transcatheter aortic valve replacement (TAVR).

Fluoroscopy is the primary tool for guidance during the procedure, but various intravascular methods may be combined with angiographic procedures. Since stent visibility is often suboptimal when X-ray fluoroscopy is used, digital image processing methods have been developed for stent enhancement. In coronary interventions, the motion of the vessels prevents the superposition of static roadmaps, and navigation within the vessels is performed by visual comparison with the displayed angiographic image and path verification by additional contrast injections during catheterization, balloon and stent placement. The dynamic coronary roadmap is a novel tool to aid navigation during the procedure, providing a real-time, dynamic overlay of the coronary tree on the fluoroscopic images used for PCI.

Post-procedure imaging varies between different imaging modalities, with or without flow reserve challenges. Patient follow-up is an essential part of clinical research. With the development of deep learning algorithms, individual follow-up tasks can be completed by AI. AI can access all data in the background to aid the physician in planning and executing the optimal patient outcome measures.

AI will have utility in analyzing the ever-growing amount of patient data being generated. Adding related technologies to assist physicians in better image analysis may significantly impact the efficiency of the interventions and patient outcomes. In the future, a critical AI enhancement will be the addition of clinical and laboratory data to the imaging data to enhance system precision further. The algorithms will improve themselves based on machine learning and deep learning algorithms.

As an emerging technology, AI will support continued innovation in interventional robotics. The future catheterization laboratory aims to enable the integration of multiple imaging modalities, online clinical decision support systems, voice-powered virtual assistants, augmented reality platforms, and automated/semi-automated robotic systems for personalized patient management, with AI-enabled technology.

future catheterisation laboratory with artificial intelligence-enabled technology.png.png

Advances at Home and Abroad.

The penetration rate of interventional robots is not high enough at home and abroad. In the vascular intervention industry, no related product has been approved by the national medical products administration (NMPA), and artificial intelligence is expected to create a pivotal breakthrough for the sector.

Applications of AI in interventional cardiology can be divided into virtual and physical branches. The virtual subfield includes informatics for machine learning (ML), deep learning (DL), natural language processing (NLP), and cognitive computing, as well as control health management systems (i.e., electronic health records and medical image analysis software) and automated clinical decision support systems. And robotic interventions represent the physical branch.

Image Interpretation: AI can assist in reconstructing, analyzing and interpreting medical images. Complete reconstruction and analysis of the entire coronary artery tree can be the input for robotically assisted procedures. HeartFlow has been approved by the FDA (Food and Drug Administration) and NICE (National Institute of Health and Clinical Excellence) for anatomic and functional assessment of coronary stenosis using non-invasive imaging with DL. DeepVessel FFR of Keya Medical (Chinese:科亚医疗) was the first Class-III AI medical device approved for clinical use by NMPA and received FDA clearance in April 2022. Siemens Healthineer has received FDA approval for TrueFusion, a cardiovascular application that integrates advanced ultrasound and angiography to improve navigation and guidance during structural heart disease interventions.

Clinical Decision Support: Clinical decision support systems (CDSS) with cognitive computing and self-learning systems that use ML, pattern recognition, and NLP to mimic human thought processes are still under development. IBM Watson Health applies cognitive technology to extract and analyze information from electronic medical records, laboratory reports, imaging reports, published medical reports, guidelines, and various Internet resources. Currently, IBM is developing Medical Sieve, an automated cognitive assistant for cardiologists and radiologists designed to aid clinical decision-making, which has enabled several cardiac imaging modalities, including the computerized detection of coronary artery stenosis in angiography.

Big Data Integration and Disease Prediction: ML has been used in cardiology to predict 1-year mortality in heart failure patients and 5-year mortality in coronary angiography datasets of patients with suspected coronary artery disease. In a pilot study, the ML algorithm was 94 percent accurate in predicting myocardial infarction in emergency department patients with chest pain. In 2020, the AI-ECG Platform developed by Lepu Medical (Chinese:乐普医疗) was approved by NMPA. Based on DL techniques and trained using tens of millions of ECG clinical big data, it has demonstrated an accuracy of 95% in automatically analyzing arrhythmia, myocardial infarction, ventricular hypertrophy, and ST-T abnormalities.

Physical Applications--Robot-assisted Surgery: The CorPath200 robotic system from Corindus Vascular Robotics, a company acquired by Siemens in 2019, is the first medical device approved by the FDA to assist in PCI procedures, helping physicians improve precision and accuracy when placing stents. In 2021, Pulse Medical (Chinese:博动医疗) released its new generation AI-QFR coronary precision diagnosis system.

Challenges

Although the number of AI-related developments in interventional procedures is exploding, and exponential growth is expected soon, many challenges must be addressed. 

challenges.png.png

Data acquisition and analysis protocols must be standardized, which in itself is a challenge since the implementation of digital standards within the clinical setting has been slow. Traditional surgery relies on the physicians' experience and subjective manipulation, which makes it impossible to be effectively standardized and hinders the development of digitalization and intelligence. Applying AI technology to extensive data analysis requires a high degree of standardization and integration of the various data sources within a hospital, between hospitals, and access to primary data from patient referrals and outcome data from outpatient follow-up. However, in most countries, each medical institute or group uses a different patient record system; the integration of imaging databases with clinical and laboratory databases within institutions is limited. In addition, traditional surgery relies on the physicians' experience and subjective manipulation, which makes it impossible to be effectively standardized and hinders the development of digital and intelligent surgery. As a result, the crucial element is to create large-quality databases with accurate and reliable annotations.

As with scientific progress, the proper application of AI depends on the right scientific questions and the precise data sources needed to answer them. Systematic biases in clinical data collection may affect AI recognition and prediction. The interpretability of AI algorithms is also a challenge. The safety and verifiability of automatic analysis and its impact on human-computer interactions may affect the utility of AI in clinical practice. Medicine is a conservative discipline, tending to apply novel methods after thorough corroboration. There must be a commitment to demonstrate accuracy through validation processes to ensure that AI can be successfully implemented into interventional practice. Routes to validate the AI-based decision processes should be explored and compared to expert decisions from the catheterization laboratory. Such algorithm verification is complicated and demanding and is a cross-disciplinary effort.

The ability to perform efficient and safe procedures remotely depends on the lag time. Hysteresis may occur due to the flexibility of interventional devices and their complex motion properties in the body. Physicians may find that the guidewire does not show as much movement on the X-ray image as it should because much of the motion is hidden by the long distance and tissue contact. What is worse, the hidden energy can suddenly be released at a certain level, increasing the risk of vascular damage. Accurately predicting and compensating for such hysteresis poses a significant challenge for AI-assisted interventional procedures.

Robotics has begun to enter the field of interventional cardiology and is now expanding toward peripheral vessels and neurovascular interventions. AI is revolutionizing the interventional therapy arena as it keeps advancing in clinical practice. Despite some initial obstacles, we are full of imagination about the prospects of artificial intelligence applied in interventional surgical robots.