Artificial Intelligence in Plastic Surgery

Artificial Intelligence in Plastic Surgery

1. Artificial intelligence in plastic surgery Moderator: Dr. Vaibhav Jain Presenter: Dr. Nishish Vishwakarma

2. Artificial intelligence (AI): simulates human cognition, interpreting and analyzing complex data to solve problems. Face recognition, Speech conversion, and Search engine recommendations.

3. The application of AI in medicine and healthcare is increasing. AI has been widely used in diabetes management, dermatology, and thoracic surgery.

4. Medical professionals must quickly and intelligently make decisions based on patient and clinical element analysis. Cognitive biases, personality factors, the doctor’s emotions, and even the environment might affect diagnosis and treatment choices.

5. Every therapeutic decision-making process involves some degree of mistake, uncertainty, and variability.  Artificial intelligence (AI) is beneficial in addressing the constraints of human behavior by providing assistance in medical decision-making and logical thinking.

6. Meanwhile, as informatization has progressed, the amount of data collected from each patient before and after surgery has increased significantly. Also, health data created by patients and off-the-shelf wearable sensors collect a lot of information. This large amount of digital data is known as "big data" . When it comes to using and analysing big data to solve hard clinical problems, independent research by hand is not enough. It is important to use AI to handle a lot of info.

7. Subfields and Concepts of AI AI is a computer program that can look at data and solve hard problems in many different situations without any help from a person.  AI creates systems or methods for analyzing information.

8. Machine learning ML is a computer algorithm that autonomously learns without specific programming. The model acquires knowledge from examples and recognizes patterns and trends in data using ML for a specific goal. As the quantity of instances and the frequency of test repetitions rise, the algorithm exhibits greater precision and predictive capability. 

9. The establishment step of ML includes data preparation and selection, training the algorithm, obtaining a series of related output results, implementation of the software and analysis, and system validation. The accuracy of the system is gradually improved with algorithm adjustment and feedback.

10. ML is divided into: supervised, unsupervised, semi-supervised, transfer learning.

11. Supervised learning is a process in which the algorithm learns from the input aspects of the data to identify and extract their properties. It then uses this knowledge to label and classify unstructured data. Supervised learning is primarily used for classification. Supervised learning is trained to observe and categorize new examples based on known instances of the category.  Predictive ML models and face recognition are supervised learning methods.

12. Unsupervised learning is mostly used to look into and describe data that already exists. It has been used to analyze molecular genetic and genomic data. Semi-supervised learning occurs between supervised and unsupervised learning because it can use labeled and unlabeled data to learn.

13. Neural Networks Neural networks employ numerous layers of calculations to mimic human behaviour to tackle complex data and derive logical inferences. NN possesses numerous "nodes". The network can achieve optimal predicting accuracy by modifying the weights of the nodes.

14. Deep learning is a category of neural networks that is distinguished by the presence of numerous layers of hidden nodes. The network has a larger scale to precisely represent complex interconnections. Deep learning algorithms have the ability to uncover the fundamental characteristics of a model, make predictions based on a large volume of data, and provide an output without the need for external supervision. 

15. The commonly used NNs are “ convolutional neural networks (CNNs)”, a type of deep learning architecture. The AI technology used to distinguish between malignant melanoma and benign nevi is based on the CNN system. CNN is an architecture within AI medicine due that is exceptionally efficient at natural signals such as speech and image processing.

16. Natural Language Processing NLP establishes a technique for understanding, interpreting, and manipulating human language. The customer service software that automatically answers questions is NLP technology. NLP can identify and extract data from a physician’s narrative documentation in electronic medical record systems.

17. Current Development Of AI i n Plastic Surgery

18. Pre-operative consultation

19. In a different study, AI looked at how people feel about plastic surgery words on social media and discovered that patients feel the most strongly about the word "liposuction." This study uses a large amount of social media data to quantitatively assess the effect of plastic surgery phrases on patient motivation and subsequent responses from the perspective of emotion.

20. It provides guidance to plastic surgeons regarding the marketing of plastic surgery from the standpoint of emotionally-driven economic decision-making. Identifying the primary area of patient interest on social media enables surgeons to more accurately anticipate and address future patient issues, effectively communicate with reliable information, provide appropriate patient educational resources, and enhance patient satisfaction.

21. During the online consultation step, artificially intelligent virtual assistants (AIVA) that use natural language processing (NLP) technology can figure out what people are trying to say and respond in the form of a conversation. An AIVA was used in a study to talk to plastic surgery patients about common issues.

22. The plastic surgery AIVA got 92.3% of the questions right, and 83.3% of the people who took it thought the answers were right. The dissemination and implementation of this technology will assist in liberating plastic surgeons from initial medical consultations.

23. Plastic surgery treatment regimens are typically customized and tailored to the patient's aesthetic preferences and the surgeon's expertise. Prior to undergoing surgery, patients typically articulate their aesthetic preferences and desired physical look. Artificial intelligence technology can be utilized to create computer-generated visuals that serve as visual aids for surgical procedures, enhancing the planning process.

24. BreastGAN is a portable tool that utilizes artificial intelligence and is fitted with a neural network algorithm. It is used to train the algorithm using breast photos of patients who have undergone bilateral augmentation. BreastGAN employs pre-operative breast photographs to automatically generate simulated results of breast augmentation.

25. An AI model can provide simulated images of rhinoplasty that align with the aesthetic preferences of the surgeon. The consistency between the generated images and the results of rhinoplasty is 92%. These technologies offer patients AI-powered forecasts to assist them in making better-informed decisions, resulting in reduced reoperation rates and enhanced postoperative satisfaction.

26. Artificial intelligence has the ability to impartially assess face attractiveness. A group of human referees assesses a set of female facial photos, and supervised classification is employed to automatically extract facial characteristics. Supervised learning employs proportion analysis through image-based assessments to evaluate facial attractiveness.

27. This technique allows for the objective evaluation of facial appearance, helping to prevent exaggerated or unnatural effects of plastic surgery.

28. Nevertheless, the implementation of this technology has the potential to diminish the variety of aesthetics, leading to a homogeneity of facial features among patients following surgery.

29. On the other hand, we need to be aware that some AI systems have negative effects on medical propaganda. Video shows in medical ads have become more trustworthy since a lot of people use photo-editing software. Images synthesized through generative adversarial network using a publicly available data set of celebrity images. These are not real human faces.

30. Deepfakes, on the other hand, are advanced AI techniques that use deep learning and computational modeling to change people's faces, facial emotions, and body movements in videos. When deepfakes were used to change a patient's video after surgery, the results of the surgery seemed out of proportion. DEEPFAKE ORIGINAL

31. Aided Diagnosis and Pre-operative Evaluation The application of AI to image diagnosis was initiated. The predominant computer-aided diagnostic technique employed in plastic surgery is the utilization of Convolutional Neural Networks (CNNs) to classify skin lesions. This system is particularly effective in distinguishing between keratinocyte carcinomas and benign seborrheic keratoses, as well as differentiating malignant melanomas from benign nevi.

32. The study demonstrated that the CNN algorithm, trained using open-source photos, exhibited superior sensitivity and specificity in identifying dermoscopic images compared to clinicians ranging from junior to chief physicians. Another multi-center prospective study demonstrated that the AI has a similar sensitivity and specificity to experts for identifying melanoma images taken with smartphones and digital single-lens reflex cameras.

33. AI has been employed in the computational examination of craniosynostosis using computed tomography (CT) images. This machine learning approach utilizes cranial suture fusion indices, as well as averages of deformation and curvature discrepancies across five cranial bones and six suture areas, as features for diagnosing craniosynostosis. The chance of right classification is 95.7%, which is similar to the accuracy of expert radiologists.

34. In the field of diagnosing uncommon medical conditions, facial analysis software now use unsupervised machine learning techniques to accurately identify abnormal craniofacial characteristics from two-dimensional pictures. These technologies aid clinicians in expediting diagnoses and minimizing errors. Moreover, the utilization of AI technology in smartphone applications enhances the convenience of disease screening for patients.

35. Surgical Decision-making and Performance During surgery, AI gives surgeons reference information to help them make decisions. It is very important to find free-flap ischemia and congestion as soon as possible during microsurgical repair of free flaps.

36. For postoperative tissue perfusion tracking, an app for smartphones was made to measure tissue perfusion based on skin colour. The algorithm knows that skin colour is a trait that can be used to mimic blood flow conditions. It was possible to recognize blood flow up to 95% of the time. This tool facilitates speedy tissue salvaging and early identification of surgical danger.

37. The AI robotic surgery system is a navigation aid for surgeons during robot-assisted surgeries. Supervised autonomous procedures have the ability to identify anatomical structures and perform surgical navigation to assist in making decisions during surgery.

38. Robotic surgical assistant technology is employed in the surgical procedure for repairing cleft lip and palate. It utilizes deep learning techniques to reduce the level of technical difficulty and enhance the results of surgical procedures.

39. In the future, there will be a development of an autonomous robotic surgical system that utilizes AI technology. This system aims to offer accurate surgical treatment, enhance surgical efficiency, minimize complications, and decrease the duration of hospital stays.

40. Video data from surgeries is being used more often because to wearable technology and head-mounted cameras. AI recognizes and extracts high-quality surgical data from videos. Surgical video analysis boosts the skills of surgeons and improves surgical methods.

41. Artificial intelligence's powerful data retrieval and processing powers significantly shorten the time it takes plastic surgeons to become knowledgeable and experienced. AI-guided autonomous surgical devices have the potential to enhance surgical outcomes in developing nations.

42. Prediction and Evaluation of Postoperative Outcomes AI builds complex models that use clinical data, images, and histopathological information to figure out the best way to treat a patient and evaluate how well the treatment worked. These models are better than traditional regression models.

43. The predictive outcome model was first used to study how wounds heal in plastic surgery. Based on the size of the wound, the patient's age, and the time between when the wound showed up and when treatment started, the formula was used to guess how fast the wound would heal.

44. An artificial neural network (ANN) study of the burn spectra taken with a reflectance spectrometer on the third day after the burn was used to predict how long it would take to heal for burn treatment and triage. 86% of the time, this method works, which is a lot better than the accuracy of a straight visual examination. The method works well enough for doctors who aren't familiar with burn systems for determining out how bad a burn is.

45. ANN is a good way to predict surgical site infections in people who had free-flap repair after head and neck cancer surgery. This method of forecast works a lot better than regular logistic regression. These predictions help formulate reasonable patient management strategies.

46. The impact of aesthetic assessment following surgery is typically very individualized for facial cosmetic procedures. AI offers an impartial assessment criterion for measuring the impact of cosmetic surgery. Artificial intelligence use deep learning techniques to extract face characteristics and utilizes convolutional neural networks (CNN) to assess both apparent age and facial beauty. These techniques provide an objective assessment of patient aesthetic improvement after surgery.

47. The algorithm indicated that orthognathic surgery made most patients (66.7%) younger and more attractive. Rhinoplasty reverses face aging by increasing facial attractiveness and decreasing perceived age relative to the patient's actual age. CNN-based AI face beauty judgments of treated cleft patients were comparable to human ratings.

48. Artificial intelligence can evaluate facial photos to determine how surgery affects emotional aesthetics. Noldus FaceReader utilizes ML to measure facial expression emotion proportions from video data. One study examined smiles in fifteen facial palsy patients after cross-facial nerve grafting and free gracilis muscle transfer. The patients' postoperative smiling videos indicated more happiness than pre-operative ones.

49. Another AI emotion recognition investigation indicated that facial paralysis patients smiled less joyfully and more negatively than others. These methods objectively measure face emotion before and after surgery.

50. AI can tell a person's gender by looking at their face. Researchers used facial recognition NN to look at how gender-typing changed in male-to-female transgender patients before and after surgery. The results showed that gender-typing changes and confidence in femininity are much better after facial feminization surgery.

51. Postoperative Monitoring and Follow-up The extensive utilization of wearable technology and mobile health technologies has increased the convenience of remote technologies for patient monitoring and communication. Surgeons assess the postoperative results by utilizing patient-generated selfies or other mobile phone monitoring applications. These technologies help to reduce the amount of time needed for follow-up. In addition, they provide surgeons with prompt follow-up data that may be converted into digital format to aid in analysis and integration.

52. LIMITATIONS AND CHALLENGES Despite AI achieving encouraging results in medicine, privacy protection and ethical issues related to its application still need to be discussed. Machine learning techniques necessitate training with large datasets in order to get satisfactory performance. Medical-related health data is extremely sensitive, and it is essential to safeguard the privacy and ownership of the data.

53. Initially, it is necessary to establish more stringent and regulated laws in order to safeguard personal information. Secondly, regulators must provide sufficient oversight of the data to prevent the disclosure of personal information. Lastly, the ethical discipline of industry personnel and the advancements in data protection technology are essential.

54. Big data needs de-identification and data sharing. Building a multi-center data platform to collect high-quality and standardized big data is the foundation for the development of AI. There must be strict requirements for informed consent, data protection, and cybersecurity. Programmers, policymakers, physicians, and patients should participate in real-world algorithmic decision-making processes to maximize fairness and respect privacy.

55. AI helps to free medical practitioners from repetitive and time-consuming simple tasks. But AI has some limitations that should be understood. First, AI can't completely take the place of doctors when it comes to evaluation and making decisions. AI is often used with conditional limits right now, and the only thing that comes out of the algorithm is an association.

56. It is important to check if the output data can be used by doctors with certain clinical patients and situations. Second, AI exacerbates pre-existing problems such as overdiagnosis, overdetection , and overtreatment. Surgeons need to fully understand AI's pros and cons to promote the application and development of AI in medicine.

57. CONCLUSION AI has been increasingly applied in healthcare settings, promoting the development of pre-operative evaluation, prognosis prediction, patient management, and postoperative monitoring. Many complex AI algorithms can learn from data and refine algorithms through learning. Physicians are also needed to optimize algorithms.

58. AI in plastic surgery is still in its infancy, and most studies have not entered clinical practice, but its advancement and prospects are promising. Plastic surgeons must master and employ these technologies to solve clinical and research issues. Future medical services may be safer and more efficient using AI.

59. Happy World Plastic Surgery Day… THANK YOU.