Publication: Modelo de predicción de la cirugía hepática laparoscópica de los segmentos 7 y 8 basado en inteligencia artificial
Authors
Herrero Muñoz, Irene
item.page.secondaryauthor
Escuela Internacional de Doctorado
item.page.director
López López, Víctor ; Robles Campos, Ricardo
Publisher
Universidad de Murcia
publication.page.editor
publication.page.department
DOI
item.page.type
info:eu-repo/semantics/doctoralThesis
Description
Abstract
INTRODUCCIÓN
La cirugía hepática laparoscópica ha avanzado notablemente en los últimos años gracias a los progresos en el conocimiento anatómico y funcional del hígado y a los avances tecnológicos. El abordaje laparoscópico ofrece claros beneficios, pero también presenta limitaciones como restricción de movimientos, una visión espacial reducida y ausencia de sensación táctil. Los resultados dependen, además, del tamaño y localización del tumor o de la extensión de la resección. La cirugía de los segmentos posterosuperiores sigue siendo uno de los mayores retos. Con el fin de estimar resultados, la inteligencia artificial (IA) se perfila como herramienta clave para generar modelos de predicción y algoritmos de riesgo personalizados que orienten el tratamiento más adecuado en cada caso.
OBJETIVOS DE LA TESIS
El objetivo de este estudio fue analizar y desarrollar un modelo de IA que demostrara utilidad para la predicción de la complejidad y resultados postoperatorios de la cirugía hepá¬tica laparoscópica de los segmentos 7 y 8, en función de determinadas características de cada paciente y de cada lesión, con el fin de ayudar en la toma de decisiones en las intervenciones de estos segmentos complejos. El objetivo secundario fue predecir qué intervenciones iban a precisar conversión a cirugía abierta.
METODOLOGÍA
Se recogieron los datos de 585 pacientes sometidos a cirugía hepática de los seg¬mentos 7 y 8 mediante abordaje mínimamente invasivo, procedentes de 19 centros especia¬lizados a nivel internacional.
La complejidad y los resultados de la cirugía se predijeron utilizando modelos de IA tras el análisis de las variables recogidas, agrupadas en función de cada uno de esos objetivos de estudio. Se analizaron distintos modelos y sus parámetros eligiendo el mejor modelo entrenado a través de un conjunto de entrenamiento y evaluado en un conjunto de prueba.
Se aplicó el método SHapley Additive exPlanations (SHAP) para facilitar la interpre-tación de dichos modelos, revelando la importancia de cada variable y su contribución a las predicciones. Se creó una aplicación interactiva para realizar predicciones de cada caso de forma individualizada, y reflejar de forma visual la contribución de las variables y su aplica¬bilidad a un escenario real.
RESULTADOS
Para realizar la predicción de la complejidad el modelo que obtuvo el mejor rendimiento tras el ajuste de hiperparámetros fue el perceptrón multicapa (multilayer perceptron –MLP–). Según este modelo, el tipo de resección es la variable más importante para la predicción de la complejidad de la cirugía, siendo las siguientes más relevantes el tamaño de la lesión más grande y el índice de comorbilidad de Charlson.
El mejor rendimiento para predecir los resultados de la cirugía se obtuvo con el mo¬delo Random Forest. En este caso, el tipo de resección y el tamaño de la lesión más grande vuelven a ser las dos variables más importantes, siendo la edad la tercera variable con mayor relevancia.
En cuanto al análisis de las variables relacionadas con el riesgo de conversión a cirugía abierta, para las variables categóricas, la localización del tumor obtuvo diferencias significati¬vas en su distribución con una mayor proporción de conversión para las lesiones del segmento 7. De las variables numéricas, mostraron diferencias las variables de pérdidas hemáticas, las complicaciones y el tiempo quirúrgico.
CONCLUSIONES
Nuestro estudio demuestra que la IA puede generar modelos predictivos precisos para la cirugía hepática laparoscópica de los segmentos 7 y 8, considerando características del paciente y de la lesión. Estos pueden emplearse para asistir en la toma de decisiones, mejorar la planificación quirúrgica, y reducir los riesgos. Además, identificamos factores que aumentan la probabilidad de conversión a cirugía abierta.
INTRODUCTION Laparoscopic liver surgery has advanced significantly in recent years due to progress in the anatomical and functional understanding of the liver as well as technological developments. The laparoscopic approach offers clear benefits but also presents limitations such as restricted movements, reduced spatial vision, and lack of tactile feedback. Outcomes also depend on factors such as tumor size and location or the extent of the hepatic resection. Surgery of the posterosuperior segments remains one of the greatest challenges. With the aim of estimating outcomes, artificial intelligence (AI) is emerging as a key tool for developing predictive models and personalized risk algorithms to guide the most appropriate treatment in each case. AIMS The objective of this study was to analyze and develop an AI model useful for predicting the complexity and postoperative outcomes of laparoscopic liver surgery of segments 7 and 8, based on specific characteristics of each patient and each lesion, to support decision-making in surgeries involving these challenging segments. The secondary objective was to predict which procedures would require conversion to open surgery. METHODS Data were collected from 585 patients who underwent minimally invasive liver surgery of segments 7 and 8, sourced from 19 specialized international centers. Surgical complexity and outcomes were predicted using AI models following the analysis of variables grouped according to each of the study’s objectives. Various models and parameters were analyzed, selecting the best-performing model trained on a training set and evaluated on a test set. The SHapley Additive exPlanations (SHAP) method was applied to enhance interpretability of the models, revealing the importance of each variable and its contribution to the predictions. An interactive application was created to make individualized case predictions and to visually reflect the contribution of the variables and their applicability in real-world clinical scenarios. RESULTS To predict surgical complexity, the best-performing model after hyperparameter tuning was the multilayer perceptron (MLP). According to this model, the type of resection was the most important variable for predicting surgical complexity, followed by the size of the largest lesion and the Charlson comorbidity index. The best performance for predicting surgical outcomes was achieved with the Random Forest model. In this case, the type of resection and the size of the largest lesion were again the two most important variables, with age being the third most relevant factor. Regarding the analysis of variables associated with the risk of conversion to open sur¬gery, among the categorical variables, tumor location showed significant differences, with a higher conversion rate for lesions located in segment 7. Among the numerical variables, blood loss, complications, and operative time showed significant differences. CONCLUSIONS Our study demonstrates that AI can generate accurate predictive models for laparosco¬pic liver surgery of segments 7 and 8, taking into account both patient and lesion characteris-tics. These models can be used to assist in clinical decision-making, improve surgical plan¬ning, and reduce risks. Additionally, we identified factors associated with a higher likelihood of conversion to open surgery.
INTRODUCTION Laparoscopic liver surgery has advanced significantly in recent years due to progress in the anatomical and functional understanding of the liver as well as technological developments. The laparoscopic approach offers clear benefits but also presents limitations such as restricted movements, reduced spatial vision, and lack of tactile feedback. Outcomes also depend on factors such as tumor size and location or the extent of the hepatic resection. Surgery of the posterosuperior segments remains one of the greatest challenges. With the aim of estimating outcomes, artificial intelligence (AI) is emerging as a key tool for developing predictive models and personalized risk algorithms to guide the most appropriate treatment in each case. AIMS The objective of this study was to analyze and develop an AI model useful for predicting the complexity and postoperative outcomes of laparoscopic liver surgery of segments 7 and 8, based on specific characteristics of each patient and each lesion, to support decision-making in surgeries involving these challenging segments. The secondary objective was to predict which procedures would require conversion to open surgery. METHODS Data were collected from 585 patients who underwent minimally invasive liver surgery of segments 7 and 8, sourced from 19 specialized international centers. Surgical complexity and outcomes were predicted using AI models following the analysis of variables grouped according to each of the study’s objectives. Various models and parameters were analyzed, selecting the best-performing model trained on a training set and evaluated on a test set. The SHapley Additive exPlanations (SHAP) method was applied to enhance interpretability of the models, revealing the importance of each variable and its contribution to the predictions. An interactive application was created to make individualized case predictions and to visually reflect the contribution of the variables and their applicability in real-world clinical scenarios. RESULTS To predict surgical complexity, the best-performing model after hyperparameter tuning was the multilayer perceptron (MLP). According to this model, the type of resection was the most important variable for predicting surgical complexity, followed by the size of the largest lesion and the Charlson comorbidity index. The best performance for predicting surgical outcomes was achieved with the Random Forest model. In this case, the type of resection and the size of the largest lesion were again the two most important variables, with age being the third most relevant factor. Regarding the analysis of variables associated with the risk of conversion to open sur¬gery, among the categorical variables, tumor location showed significant differences, with a higher conversion rate for lesions located in segment 7. Among the numerical variables, blood loss, complications, and operative time showed significant differences. CONCLUSIONS Our study demonstrates that AI can generate accurate predictive models for laparosco¬pic liver surgery of segments 7 and 8, taking into account both patient and lesion characteris-tics. These models can be used to assist in clinical decision-making, improve surgical plan¬ning, and reduce risks. Additionally, we identified factors associated with a higher likelihood of conversion to open surgery.
publication.page.subject
Citation
item.page.embargo
Collections
Ir a Estadísticas
Este ítem está sujeto a una licencia Creative Commons. http://creativecommons.org/licenses/by-nc-nd/4.0/