Page 111 - Read Online
P. 111

Page 153                                                              Shen et al. Art Int Surg. 2025;5:150-9  https://dx.doi.org/10.20517/ais.2024.71

               the benefits of AR and establish its efficacy in different clinical settings.


               AI APPLICATIONS INSIDE THE OPERATING ROOM
               AI is revolutionizing the use of intraoperative imaging and visualization, particularly in autologous breast
               reconstruction, by generating AR models based on the data collected through advanced imaging
               technologies. These advanced imaging technologies include Near-Infrared Laser-Assisted Indocyanine
               Green (ICG) imaging, which has emerged as a reliable assessment tool of direct perforator perfusion zones.
               From raw ICG data, the SPY-Q Analysis Toolkit (Novadaq Technologies, Inc., Toronto, Canada, 2009) can
               calculate quantifiable parameters like relative and absolute perfusion values, so surgeons can objectively
               identify the dominant perforators and their perfusion zones. Sacks et al. demonstrated success in
               conventional free anterolateral thigh flap dissection with the aid of laser-assisted ICG, with only 1 flap loss
               in 15 due to venous congestion . Ongoing efforts have shown early success in developing high-fidelity AI
                                          [14]
               models using ICG fluorescence angiography to assist in flap trimming decisions intraoperatively.
               Singaravelu et al. developed an ensemble subspace k-nearest neighbor model that was able to predict the
                                                  [15]
               optimal excision area with 99.3% accuracy .

               Building upon SPY-Q-guided manual identification, the integration of 3D-printed vascular modeling
               (3DVM) enables the visualization of more spatial information with depth, orientation, and direct tactile
               feedback in DIEP flap breast reconstructions . Following these advancements in 3DVM, Meier et al.
                                                       [16]
               introduced a novel approach using projected AR for perforator mapping in DIEP flap breast reconstruction
               with the aim of increasing interactive visualization of vascular anatomy . The study utilized a self-aligning
                                                                           [17]
               projection device equipped with a thermal camera, which projected a color-coded thermal map directly
               onto the patient’s abdomen. This system employed dynamic infrared thermography (DIRT) to identify
               perforators based on their heat signatures, offering a quick and non-invasive method for perforator
               mapping. While the technology is still nascent, early results are promising, with the majority of
               DIRT-identified perforators also identified on preoperative Doppler and CTA, as well as intraoperatively, all
               without the need for radiation or contrast.


               Recent advancements in AI have led to the development of a novel approach using DIRT coupled with
               CNNs for automated blood vessel detection in DIEP flap breast reconstruction . These deep learning
                                                                                      [2]
               models can extract complex features from medical images, capturing both local and global context to
               enhance the accuracy of perforator identification and flap perfusion assessment .
                                                                                  [18]

               The study by De La Hoz et al. utilized a DIRT protocol to capture images of the abdominal area after
               cooling , generating a dataset that was used to train and test the CNN. The CNN was designed to directly
                     [19]
               segment these thermograms to identify perforators with high confidence. Saxena et al. further advanced the
                                                                                     [20]
               field by exploring deep learning for vessel segmentation in DIEP flap planning . Their approach used
               rectus abdominis muscle segmentation from computed tomography (CT) scans to select the appropriate
               region of interest for targeted vessel segmentation and diameter measurement. However, the model’s
               reliance on synthetic datasets requires further validation.

               A separate study by Seth et al. segmented CTA data using AI-enabled software, Materialise Mimics
               (Materialise NV, Belgium, 2017) . Materialise Mimics can provide automated segmentation of anatomical
                                           [21]
               structures on CTA using proprietary AI that involves image thresholding, ML, deep learning, and graph
               learning algorithms applied to computer vision. This study created a virtual overlay of 3D models
               containing color-coded representations of patient anatomy, subsequently manually aligned with the patient
               using fixed anatomical landmarks . This allowed for the visualization of DIEP’s extra- and intramuscular
                                            [22]
   106   107   108   109   110   111   112   113   114   115   116