Page 110 - Read Online

Shen et al. Art Int Surg. 2025;5:150-9 https://dx.doi.org/10.20517/ais.2024.71 Page 152

Predictive analytics powered by ML models have emerged as crucial tools in preoperative risk assessment
and management. In a study by O’Neill et al., a ML model was used to predict flap failure in microvascular
breast reconstruction. It identified high-risk patient groups, such as those with obesity, comorbidities, and
[4]
smoking habits, allowing for targeted interventions and improved outcomes . With tools like this, surgeons
can immediately stratify the risk of each patient and provide personalized counseling to high-risk patients
on lifestyle modifications to prevent flap failure.

The Artificial Intelligence-Based Multimodal Risk Assessment Model for Surgical Site Infection (AMRAMS)
[5]
exemplifies the potential of AI in risk management . By incorporating demographics, preoperative lab
results, and free-text preoperative notes, AMRAMS leveraged deep learning methods, including CNNs, to
enhance accuracy. Compared with the National Nosocomial Infections Surveillance risk index manually
scored by surgeons, AMRAMS offered significantly higher accuracy. The inclusion of semantic embeddings
of preoperative notes further improves model performance, suggesting that AMRAMS could potentially
replace traditional risk indices to provide highly personalized guidance for preoperative interventions.

AI-aided surgical planning is also transforming preoperative consultations. Traditional preoperative
consultations often involve the use of diagrams, photographs, and verbal descriptions, which may not
effectively render the potential outcomes of the surgery. In a study published by Chartier et al., BreastGAN,
an AI tool driven by generative adversarial neural networks (GANs), was able to simulate breast
[6]
augmentation outcomes based on preoperative images . The use of another complementary technology in
tandem with AI is particularly revolutionizing this space. Augmented reality (AR) overlays digital content
onto the real world. When combined with AI that provides computational analysis and predictive
capabilities, AR creates more immersive and comprehensive ways to tangibly visualize surgical outcomes,
helping center patient expectations and improve postoperative satisfaction.

A prospective study with patients undergoing breast augmentation reported that patients were satisfied with
preoperative 3D simulation using Arbrea Breast Software (Arbrea Labs, Zurich, Switzerland, 2018) and
postoperative outcomes, measured with a visual satisfaction analog scale and BREAST-Q Augmentation
[7]
module . Arbrea, similar to BreastGAN, uses a type of generative AI called GANs in combination with AR
and 3D simulation to help patients visualize surgical outcomes. Additionally, in another cohort of 40
patients undergoing breast reconstruction, virtual reality (VR) tools (3D imaging) and external sizers were
[8]
shown to be the most effective among the five methods tested in choosing the implant volume .
Furthermore, the Crisalix portable 3D surface imagers, driven by deep learning, predicted breast volume
with accuracy closely matching estimates from experienced plastic surgeons and actual intraoperative
specimen weights .
[9]
Despite these exciting results, there are several barriers hindering the wider adoption of AI and AR in
preoperative consultations. Users report that AR tools like the 3D HoloLens can be cumbersome and
challenging to use . In terms of breast volume prediction, discrepancies in accuracy may exist between
[10]
different AI softwares, highlighting the need for additional testing and head-to-head comparison when
choosing the application carefully . Moreover, the cost of implementing AR technology for preoperative
[11]
planning remains unclear, largely due to its scarce implementation thus far . However, the cost of AR
[12]
modalities that may implement AI is relatively fixed, such as the Microsoft 3D HoloLens, which retails for
about $3,500; thus, their gradually decreasing cost per use may be beneficial for long-term
implementation . Effective utilization also requires specialized training for surgeons. Therefore, future
[13]
studies should focus on addressing these limitations by exploring cost-effective solutions and developing
standardized training protocols. Further, large-scale randomized controlled trials are necessary to validate