Saltar para o conteúdo principal

FCCNFCT Digital Services

Access the national information on scientific and technological funding in Portugal

Activity Funded

Optimization of Surgical Workflows and Patient Outcomes Using Artificial Intelligence

Optimização da Alocação de Recursos em Cirurgia e Melhoria doTratamento de Pacientes Usando IA

Reference
2024.07248.IACDC
Project Start Date
2025-04-01
Project End Date
2026-01-31
Principal Investigator
Scientific Area
Exact sciences
Funding Program
Inteligência Artificial, Ciência dos Dados e Cibersegurança de relevância na Administração Pública

Abstract

Healthcare services are progressively moving towards more personalized and less invasive procedures [1]. This minimizes surgical trauma and results in faster recovery periods with earlier discharge. Therefore, it comes as no surprise that minimally invasive surgical procedures are becoming standard in a variety of abdominal conditions, such as cholecystectomy, colorectal surgery, and appendectomy [1]. Contrary to open surgery, minimally invasive surgery only requires the making of 3 to 5 small port incisions no bigger than 1cm. One port allows the insertion of a videoendoscope to guide the surgical team, while the remaining ports enable the insertion of medical instruments. The usage of small cavities significantly reduces the post-operative pain, speeds-up recovery time, and avoids unnecessary exposition of the internal body to the environment and potential infections. Current videoendoscope devices convey high quality images, magnifying the field of view up to ten times . This makes it possible for a surgeon to clearly inspect the anatomical structures and difficult-to-access cavities, resulting in more precise interventions. The use of digital devices also allows for the recording of entire procedures, which has led to the emergent fields of surgical AI (SAI) [2,3] and surgomics [4]. The former addresses various problems: i) detection of surgical instruments and/or anatomical structures [5]; ii) action recognition [6]; and iii) identification of surgical phases [7]; while the latter aims to identify biomarkers for predicting patient outcomes but is still in its inception. SAI addresses challenging problems, but these fail clinical translation , as critical points are not being considered: i) prediction of surgery duration time and potential adverse events, both pre and during surgery; ii) prediction of patient outcomes after surgery (e.g, status at 30 days); and iii) identification of ideal/incorrect usage of surgical instruments for surgeon feedback and training. i) and ii) are fundamental for patient-centric healthcare. Additionally, both play a tremendous role in hospital management, as longer than expected surgeries and patient stays incur higher costs for the hospitals and result in the increase of the surgery waiting lists – both challenges that are currently faced by the Portuguese National Healthcare System (SNS) and worldwide [8,9]. The third point pertains to surgeon expertise, as it is commonly accepted by the medical community that poorer surgical outcomes are directly associated to the quality of the surgery [10]. Therefore, it would be ideal for a SAI system to be able to provide feedback to surgeons regarding their skills, guaranteeing that they could keep improving their performance over time, as well as speeding up the training of surgery interns, increasing the hospitals’ human resources. The OptSurgAI project aims to bring AI models closer to clinical translation, through the development of two prototypes : a) an AI system to estimate the duration of laparoscopic procedures before and during surgery; b) an AI system to predict patient outcomes that identifies the probability of adverse events in a 30-day window. Standard SAI methods use only videos acquired during surgery. However, we advocate that this is not enough for a personalized approach, as all information about the patient is being disregarded [4]. Thus, we propose to leverage the multimodal information that is already available at the hospitals to achieve more personalized estimates. We will develop multimodal AI prototypes combining: 1) radiology (CTs and ultrasound acquired prior to surgery); 2) patient metadata (demographics, comorbidities, medication, etc); and 3) surgical video. Given the impact of surgeon skill on patient outcomes, we aim to lay the foundations for an AI approach to be used for surgeon training. We will develop an innovative methodology based on affordances [11,12] to capture the interactions between surgical instruments and anatomical structures , which will act as a proxy to evaluate the skill of a surgeon. Our prototypes will contribute to improving the overall hospital management, allowing a more informed allocation of resources (therapeutic and human). Moreover, our work will help signal risk patients that require personalized treatments and support the continuous training of surgery interns and more advanced physicians. Finally, we plan to publicly release all data acquired during the project in an open-source platform to foster future investment on optimizing surgical procedures. OptSurgAI is led by a multidisciplinary team of AI researchers from IST-ID , with recognized expertise on medical image analysis and affordance models, and general surgeons from HFF . Our teams already have an established collaboration [13]. Thus, we are in the right place to successfully conduct this project.

Institutions

Main Institutions

  • Associação do Instituto Superior Técnico para a Investigação e Desenvolvimento (IST-ID)

Other Institutions

  • Hospital Professor Doutor Fernando Fonseca EPE (HFF)
  • Associação para a Investigação e Desenvolvimento da Faculdade de Medicina (AIDFM)

Funding 124.999,38 €

Fundação para a Ciência e a Tecnologia (FCT) - Portugal

0,00 €

União Europeia - Estrutura de Missão Recuperar Portugal (UE - EMRP)

124.999,38 €
  • 2025-04-01 a 2026-01-31
  • 2024.07248.IACDC