Pancreatic ductal adenocarcinoma (PDAC) is one of the main causes of cancer-related mortality worldwide [1]. The choice of surgical treatment is based on the location of the tumor, and the involvement of other structures. The clinical presentation differs according to the site of the tumor, but PDAC is mainly an asymptomatic tumor, and when symptoms occur, the disease is usually advanced. For these reasons, developing strategies for early diagnosis and treatment is paramount. Technology has proved an excellent tool over the years for the early diagnosis of this pathology and decision-making regarding a surgical approach. Artificial intelligence (AI) has been in development for decades, and now plays an important role in medicine, including pancreatic surgery.

This Editorial mainly focuses on the current role and future potential perspectives of AI applied in pancreatic surgery.

Technology has been an ally of surgeons for years, and the rapidly evolving tools available make a better treatment possible for our patients. One of the most important tools is AI, defined as the ability of a machine to imitate intelligent human behavior, developing algorithms that can execute cognitive tasks [2]. This is a constant and rapidly growing field, with potential applications in all aspects of medicine [3].

The number of publications regarding AI has massively increased over the last 20 years: a 40× increase has been registered from around 1,000 in 2000, to almost 40,000 publications in 2022, as seen in Fig. 1.

Fig 1. Number of artificial intelligence (AI) publications in PubMed.

The use of AI in medicine has been incrementally creating more and more potential opportunities for a personalized approach. What started in the 1950s as a series of “if, then” rules can now use models for the diagnosis and prediction of therapeutic response [4]. However, despite numerous advantages, the rapid development of AI also entails ethical questions and potential issues to be solved, such as the need for protection of patient data, and possible risks of inaccurate results (false positive and false negative). Additionally, when an AI-controlled medical device fails, it will be paramount to determine who will be responsible [2].

AI: preoperative applications

During the preparation for a surgical operation, AI is used in the following scenarios:

■ AI can analyze the simulated performance of a surgeon and provide numerical/quantitative scores. An example is the well-known and well-used SimNow^TM simulation software (Intuitive Surgical Operations, Inc.), which uses the da Vinci Xi surgeon console and computer software to train on 33 exercises to improve surgical skills [5].

■ The anatomy of the pancreas may be analyzed using 3D printing for reconstruction based on non-contrast computed tomography (CT) images, giving the opportunity to better understand the relationship between pancreatic parenchyma and vascular structures [6].

■ Deep learning and neural networks play an important role here for a decision on the most adequate treatment, especially in the case of a tumor with borderline resectability and/or that is locally advanced. Synapse VincentTM (FUJIFILM Corporation) is a software designed using deep learning algorithms. Its function is to extract 3D images from CT scans, and create simulations of pancreatic parenchyma, the pancreatic duct, and vascular anatomy. This has a potentially important role for surgical planning [7].

■ Radiomics is the term used for the identification of innumerable features using digital images and locations known as the region of interest (ROI). This information is then used for the prediction of later events, such as postoperative morbidity and mortality. Radiomics uses AI combined with medical images to allow the analysis of gray level patterns, which are usually not detectable by the human eye, correlating this with the texture of the tissues. This textural analysis, obtained by CT or MRI, allows the extraction of data without invasive procedures, reducing cost, time, and risk for the patients [8,9]. Another use described is the use of machine learning based textural analysis from non-contrast CT scans to estimate the parenchymal characteristics of the pancreas, and predict postoperative pancreatic fistulae after pancreatoduodenectomy. This was studied creating algorithms using databases with manually graded pancreatic hardness by a single surgeon, comparing this data with post operatory outcomes and histopathology [10].

AI: intraoperative applications

The development of minimally invasive pancreatic surgery makes a large amount of intraoperative data available for further processing with AI applications (e.g., surgical videos).

During the intraoperative process, AI may be integrated under these situations:

■ Surgomics is defined as the process of gathering the characteristics of a surgical procedure (derived from intraoperative data) to qualify processes in the operating room (OR). Analyzing intraoperative data and using it together with preoperative features (from clinical data and Radiomics) may be able to predict postoperative morbidity and mortality, as well as provide feedback for surgeons, and even inform the surgeon of what other surgeons would do in their situation [11].

■ AI enables surgeons to see, in real time (augmented reality [AR]), tumors and their relationships to intra-parenchymal vascular structures using 3D AR intraoperative images (SmartLiver^TM) created from 2D images. Solid organs are currently the best intra- abdominal organs to integrate with AR, due to the amount of data available from 2D images [12].

■ Cross-AI systems provide landmark detection during laparoscopic procedures, especially cholecystectomies, as described by Fujinaga et al. [13]. Deep Learning is used along with laparoscopic cholecystectomy video datasets, creating learning models that recognize the different surgical phases and anatomical structures. This helps surgeons prevent errors by protecting key structures from lesions determining “no fly zones” (dissection areas that should be avoided).

■ The use of AI along with AR during trocar placement helps to better identify safety landmarks in real-time. This technique is due to the integration of deep learning with a laparoscopic instrument and alarm system. The analysis of surgical videos showing the use of Optiview trocar placement helps recognize the different abdominal wall layers, leading surgeons to a safe approach [14].

AI: postoperative applications

The applications of AI in the postoperative context may be summarized as follow:

■ Prediction of complications using the American College of Surgeons National Surgical Quality Improvement Program (ACS−NSQIP) has been compared with the use of machine learning on large clinical datasets. Machine learning algorithms can improve predictive accuracy by capturing complex, nonlinear relationships observed in the data, using tree model decisions to compare the 30-day prediction of mortality among patients with low estimated morbidity and mortality [4].

■ Prediction of postoperative pancreatic fistulae using AI algorithms created from 16 risk factors, such as clinicopathological characteristics, imaging, intraoperative findings, BMI, preoperative serum albumin, and fluid infusion [15].

■ Prediction of the occurrence of postoperative complications within 30 days, including stroke, wound dehiscence, cardiac arrest, renal failure, and surgical site infection, as described by Merath et al. [4].

■ Prediction of early distant recurrence (less than 12 months after surgery) in PDAC, identifying variables such as tumor necrosis on imaging evaluation, radiomic feature, and CA19-9 level, using a model described by Palumbo et al. [16].

Many optimizations can be made in the OR environment using AI to control, for example, lighting, temperature, supply management, or to integrate diagnostic tools, and to accomplish automatic robotic docking. The safety of the procedure can be improved with AI-based decision-making assistance or AI warnings (e.g., missing steps or risk of iatrogenic injury) when a mistake is about to occur.

Although AI in active surveillance has been used in prostate cancer and has revealed promising results, it is still not routinely used in pancreatic cancer [17]. While AI can recognize different phases of a surgical procedure, it is still not able to fully and reliably reproduce them. Even though AR is being tested for liver surgery, it is still not efficient enough to be used in pancreatic surgery.

Recently, indocyanine green (ICG) has been used for neuroendocrine pancreatic tumor recognition. These lesions are highly vascularized, and the combined use of ICG and AI may increase the sensitivity of intraoperative tumor recognition [18].

Although autonomous surgery is being developed, there is still much work to do. A computer program generates a plan to complete complex surgical tasks on deformable soft tissue, such as suturing and intestinal anastomosis, or autonomous grasping [19]. The robot must discern tissues and instruments, and pinpoint their 3D spatial locations using only endoscopic imagery. While at present, it is possible to perform simple dissection using AI, which would be promising for pancreatic surgery, it is difficult to rely on an entirely autonomous robot for complete procedures; the challenge lies in ensuring safety and achieving submillimeter precision within a compact workspace, relying heavily on accurate visual feedback from both kinematics and computer vision [20].

It has been proven that with the constant development of new technologies, including new sensors and techniques of data processing, the capabilities of AI in pancreatic surgery provide unlimited possibilities, and are growing exponentially.

None.

No potential conflict of interest relevant to this article was reported.

Conceptualization: AD, AM, PCG. Data curation: AD, AM, LB, JC, PL. Methodology: AD, PCG. Visualization: AD, AM. Writing - original draft: AD, AM, PCG. Writing - review & editing: AD, AM, PCG.