Use of regional anaesthesia (RA) and peripheral nerve block (PNB) is growing, although general anaesthesia (GA) is still more common in general surgical practice. Around 65% of all procedures amenable to a regional technique currently use GA, and current UK National Institute of Health and Clinical Excellence (NICE) guidance is that all regional anaesthesia should be performed using ultrasound guidance. However, further increases in regional anaesthesia are expected, as there are significant patient and economic benefits. In particular, the per-procedure costs of regional anaesthesia are considerably less than for general anaesthesia.
Although growing, ultrasound-guided regional anaesthesia is difficult to learn and difficult to perform. There are significant hand-eye-coordination issues as the clinician must simultaneously manipulate both the needle and the ultrasound probe in order to guide the needle to the target. In addition, both the needle and target anatomy can be very difficult to see on the ultrasound image.
The investigators believe that a computer-aided system that highlights key anatomical features on the ultrasound image would make this procedure safer for the patients and simpler for the clinician.
Currently, the leading method for automatic image segmentation uses deep learning, for which many thousands of training images are required. There have been several successes in applying these techniques to medical images, including ultrasound. However, it appears that relatively little attention has been given to automatic segmentation of ultrasound images for regional anaesthesia.
The closest reference to our proposed research describe a method to locate the median nerve in ultrasound images of the forearm. There has also been a Kaggle challenge to segment the brachial plexus nerves in the neck. Multiple anatomical regions can also be segmented at the same time. However, none of these studies are directly applicable to clinical use as they deal only with images captured from healthy volunteers. Neither do they consider how these techniques could be used to aid anaesthetists performing regional anaesthesia in the clinic.
The rationale for this study is to determine whether real-time automatic highlighting of key anatomical features can help clinicians perform ultrasound-guided regional anaesthesia.
Medaphor's proposed system for automatic highlighting uses deep learning and requires many thousands of images for the system to learn from. However, machine learning algorithms such as deep-learning are highly dependent on the images used to construct them. In particular, care must be taken to ensure these algorithms are trained using images representative of those the algorithm will encounter when in use.
A key part of regional anaesthesia involves injecting anaesthetic into the space near the relevant nerve. Once introduced, both the needle and anaesthetic can be seen on the ultrasound image. Models trained using non-invasive images recorded from healthy volunteers will not be sufficient. Although non-invasive images from volunteers may contain the key anatomical features, they will not show the needle and anaesthetic.
The simplest option to capture representative images is to select patients who will be having regional anaesthesia and record the image data directly from the ultrasound machine as the procedure is performed. This method of recording is transparent to both clinician and patient, and does not affect the patient's treatment in any way.
In addition, the ultrasound machines used for regional anaesthesia are not connected to PACS and no patient identifiable information is entered into the ultrasound machine. Since the ultrasound image only is recorded, the videos are completely anonymous and cannot be traced back to any individual.
This study is part of a larger research programme and will build and validate a system capable of highlighting the key anatomical features. Once the system is complete, a further study will be conducted to test it in the clinic to determine potential benefits to patients and clinicians.
