The identification of the neural structures, particularly by inexperienced trainees of regional anesthesiology, may appear a major problem. As pointed out earlier, the knowledge of sonoanatomy is a condition needed and indispensable to use ultrasonography in regional block procedures, the popularity of which is growing each year [
11]. It has also been observed that the interpretation of AI-supported images may be one of the methods to improve the success rate for the procedure of nerves or nerve plexus blocks [
10]. The “head-up-display” which shows the prompts in real time coding different anatomical structures with specific colors may ensure a shorter time needed to identify the neural structures and to confirm that further decisions made by the operator are correct [
13]. The technology is now rapidly developing, including software capable not only of the recognition of anatomical structures but also tracing the needle in human tissues.
2.1. U-Net Architecture and BPSegData
Discussions on the recognition of anatomical structures by USG systems comprising a computer with appropriate software and peripheries, i.e., ultrasound heads, must not miss the notions of machine learning (ML) and deep learning (DL) and, in particular, deep convolutional neural networks (DCNNs). To make things simple, the notions refer to the commonly available IT tools which were originally developed as mathematical concepts. One of those, used earlier for the segmentation of biomedical images and later in medicine, is the U-Net architecture developed in 2015 by Olaf Ronneberger of the Computer Science Department and BIOSS Center for Biological Signaling Studies, the University of Freiburg, Germany. It offers the possibility to work on images in shades of gray, including the diagnostic radiological images, while the main task is the segregation and segmentation of images with potent coding and decoding solutions.
Such a breakthrough and precise technologies capable of differentiating organs and other anatomical structures through the pixel analysis enhance the clinicist’s efficiency who assesses the diagnostic images in a classical way. U-net will precisely probe among the most important development pathways for the identification and classification of anatomical structures [
14,
15]. The clinical studies carried out by the Department of Experimental Surgery, McGill University, Montreal, Quebec, Canada, evaluated the usability of the U-net architecture in the context of identifying the interfascial space TAP, and those conducted in the hospital in Hässleholm, Sweden, evaluated the localization of the femoral nerve, showing that this technology may offer great support, in particular, to less experienced practitioners (
Figure 1 and
Figure 2).
Results from a large sample, i.e., 25,000 images for the TAP space and 1410 for the femoral nerve, obtained within the so-called big data framework, confirmed the effectiveness of the method offered by the U-Net, at the level of 73.3% and 74%, respectively [
10,
11]. Another study, which compared the effectiveness of the deep learning scheme with manual segmentation performed by a physician for the visualization of the brachial plexus, proved, on the basis of the gathered data (BPSegData—
Figure 3), the effectiveness of the identification of that structure at the level of approximately 50% [
16].
Today there are several platforms based on machine learning technology at different levels of development which may offer significant support to medical practitioners in the area of regional anesthesiology, the most important of which include the following listed below.
Nerveblox—a technology supplied by Smart Alfa Teknoloji San. Ve Tic. A.S., Ankara, Turkey, is a software implemented in 2020 to convert the ultrasound images downloaded in real time from the ultrasound system to the computer, presenting then the anatomical structures, e.g., muscles, pleura, arterial and venous vessels, bone structures and, of course, nerves and nerve plexuses in the given colors (
Figure 4).
Another example to prove the proper prediction of a particular anatomical structure is an USG image, presented in
Figure 5, reflecting the pectoralis major, the pectoralis minor as well as the subclavian artery and vein.
As illustrated in
Figure 5, the Nerveblox technology may also be used in regional anesthesia to identify the interfascial space, while
Figure 6 below indicates, in yellow, the spots to deposit the local anesthetic for PECS I and II blocks.
The Nerveblox is capable of recognizing the USG images in real time thanks to the abovementioned mathematical algorithms referred to as the convolutional neural networks (ConvNets). The potential of neural networks is immense, yet their capabilities are now used on a marginal scale. The technology itself is based on the detection of highly complicated relations between the captured images and those stored by the device (Nerveblox, in this very case), as well as the correlation between the adjacent pixels. In the case of accordance, the computer automatically confirms the case and assigns particular colors to the individual anatomical structures [
18,
19]. At present, Nerveblox is dedicated to 12 regional blocks, the details of which are illustrated in
Figure 7.
ScanNav Anatomy Peripheral Nerve Block and NeedleTrainer (software version 2.2) is a system supplied by Intelligent Ultrasound, Cardiff, UK. Also supported by artificial intelligence while scanning the given regions of the patient’s body, it creates a colorful overlay on the monitor to indicate the requested anatomical structures. Similar to Nerveblox, the device makes use of deep learning based on the U-Net architecture. Here, the so-called big data is a database containing 800,000 models of USG images which make the points of reference for particular anatomical structures, in consequence creating a colorful overlay, as presented in
Figure 8 [
15,
20,
21].
ScanNav Anatomy PNB presents the ten most common types of regional block (
Figure 9). They are the following:
Axillary-level brachial plexus;
Erector spinae plane;
Interscalene-level brachial plexus;
Popliteal-level sciatic nerve;
Rectus sheath plane;
Sub-sartorial femoral triangle/adductor canal;
Superior trunk of brachial plexus;
Supraclavicular-level brachial plexus;
Longitudinal suprainguinal fascia iliaca plane.
Figure 9.
Examples of artificial intelligence color overlay for each peripheral nerve block studied (from [
21]).
Figure 9.
Examples of artificial intelligence color overlay for each peripheral nerve block studied (from [
21]).
The appropriate selection of a regional block is confirmed by the fact that 7 out of 10 examples specified in the so-called Plan A Regional Anesthesiology (the basic teaching standard—
Table 1) are contained in the ScanNav Anatomy software.
It is also worth mentioning that the system cooperates with classical USG system devices furnished with digital sockets to transmit images—HDMI or DVI. Note also that in October 2022, ScanNav Anatomy was approved by the American Food and Drug Administration (FDA). The system has been qualified within the framework of a de novo program which effectuated the formation of a new category of devices for the visualization and color coding of anatomical structures used in AI-supported regional anesthesia [
22,
23].
Apart from the color-coded identification of anatomical structures, the system is also capable of simultaneous presentations of instruction videos showing techniques used to accomplish a selected procedure of regional anesthesia which makes it an excellent training tool for any anesthesiologist performing regional anesthesia (
Figure 10).
The latest function furnishing ScanNav Anatomy is the NeedleTrainer which makes use of retractable needles for an instant invasive simulation of needling in human tissue. The major advantage of this solution is the possibility of developing the skills, in particular “hand–eye” coordination, during any invasive procedures including the regional block (
Figure 11). However, the technology itself is not a haptic solution and does not fully reflect the block condition, appearing as a challenge for manufacturers [
24].
The clinical studies carried out so far with the use of ScanNav Anatomy have clearly justified the application of the technology to support trainees as well as expert educators in the area of regional anesthesia [
25,
26]. This thesis may be confirmed by the results collected and described in April 2021, following a clinical study carried out in several British hospitals. During the study, experts were asked a number of questions. The study made use of the Delphi surveying method where experts highly valued the ability of ScanNav to highlight the anatomical structures of key importance for the block, and the highest scores where achieved for imaging the rea of the adductor canal, the brachial plexus in the area of the axillary fossa and the iliac fascia superior to the inguinal ligament. In the experts’ opinion, the applicability of the software to recognize individual critical structures to accomplish a technically correct block ranged between 95% and 100%, while highlighting the localization target was assessed at the level of 100% in the case of 31 anatomical structures out of 34 evaluated ones. The experts emphasized that in five cases out of the seven evaluated areas (i.e., the plane of the spinal extensor, the rectus abdominis sheath, the iliac fascia above the inguinal ligament, the adductor canal and the sciatic nerve in the area of the popliteal fossa), ScanNav would ensure a 100% benefit for less experienced practitioners and the medical trainees when confirming the correct location and positioning of the ultrasound head, while in the remaining two areas (imaging of the brachial plexus in the axillary fossa and the supraclavicular region), the score amounted to 97.5% [
20].