Page 25 - Read Online
P. 25

Allam et al. Plast Aesthet Res 2024;11:19  https://dx.doi.org/10.20517/2347-9264.2024.21  Page 5 of 16

               the DIEA perforated the anterior fascia matched perfectly with the virtual overlays. The time required for
               perforator tracking with AR was comparable to Doppler examinations, underscoring AR's potential utility
                                [12]
               in surgical planning .

               Assessing AR's clinical utility, Hummelink et al. found that AR projections [Figure 2B] significantly reduced
               operative time and improved the accuracy of perforator identification compared to Doppler ultrasound .
                                                                                                       [17]
               The study reported a reduction of over 17 min for perforator mapping and 19 min in flap harvesting time in
               the AR group. Additionally, the AR model correctly identified 61.7% of perforators, compared to 41.2% in
                                                                                                       [17]
               the Doppler group. However, the study lacked sufficient power to detect differences in complication rates .
               Current research demonstrates the proof-of-concept and non-inferiority of AR-assisted perforator
               identification and selection in autologous breast reconstruction [12,16,17] . Future studies are imperative to
               further  explore  AR's  impact  on  clinical  outcomes  and  assess  its  cost-effectiveness  for  broader
               implementation in routine breast microsurgical practice.


               Artificial intelligence
               Artificial Intelligence (AI) in surgical planning is a rapidly evolving and promising technological
               advancement in breast reconstruction. One example of its powerful application is its use in preoperative
               imaging. Given that imaging analysis for perforator selection is both labor-intensive and time-consuming,
               integrating AI and machine learning processes such as convolutional or deep neural networks into this
               process presents a viable solution. This integration could enhance time and cost efficiency while minimizing
               observer bias and human error in image interpretation.

               Traditionally, MRA and CTA analyses for perforator selection rely heavily on the expertise of technicians
               and radiologists. The incorporation of AI into this process could streamline preoperative planning and
               reduce the likelihood of error. A prospective study involving 40 patients undergoing DIEP flap
               reconstruction demonstrated that semi-automatic identification of perforators from CTA images, facilitated
               by computer vision AI and machine learning, saved approximately two h of preoperative planning time
               compared to manual analysis . While the number of localized perforators was similar in both approaches,
                                        [18]
               the AI technique showed reduced error in perforators larger than 1.5 mm in caliber. However, it also
               indicated a slight increase in error for smaller perforators, though these discrepancies were not clinically
               significant and did not impact surgical technique .
                                                        [18]
               The utility of AI extends beyond perforator selection, offering the potential for automated, preemptive
               screening of patients for significant postoperative complications. Flap failure, while rare, is a significant
               complication of autologous breast reconstruction and is challenging to predict due to the absence of clear
               individual risk factors . O’Neill et al. developed a multifactorial ML algorithm that stratifies various patient
                                 [19]
               and treatment factors into risk categories for flap failure. Analyzing a retrospective cohort of 1,012 DIEP
               flap breast reconstructions, the model identified four high-risk subgroups based on BMI, comorbidities, and
               surgical history with high specificity for predicting flap failure . While the study size was smaller than
                                                                      [19]
               typically recommended for high predictive power in ML, resulting in lower-than-expected sensitivity, it lays
               the groundwork for future research in applications of AI . This represents a proof-of-concept for an ML
                                                                [19]
               program that could screen high-risk patients for flap failure, which in turn could allow surgeons to
               pre-emptively consider alternative reconstruction options.


               Looking ahead, with adequate preoperative and postoperative data, the combination of established methods
               with ML and computer vision AI could be harnessed to analyze the characteristics of perforators and
   20   21   22   23   24   25   26   27   28   29   30