Publicación: Automatización de codificación y resumen de informes de exploraciones radiológicas de próstata
Cargando...
Fecha
2023
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (España). Escuela Técnica Superior de Ingeniería Informática. Departamento de Lenguajes y Sistemas Informáticos
Resumen
La radiología constituye uno de los pilares fundamentales de la medicina hoy en día al apoyar tanto el diagnóstico, como el tratamiento. El informe radiológico, a su vez, es un componente esencial del estudio en el que se fundamenta la interpretación de los hallazgos por parte del médico remitente. La introducción de registros electrónicos de salud y la digitalización de la información acumulada por el sistema sanitario han creado una necesidad de estructuración de la información contenida en formato de texto libre en los informes clínicos para habilitar su aprovechamiento tanto en la práctica clínica como en la investigación médica. Con el fin de contribuir tanto a la calidad de la comunicación entre los radiólogos y los médicos remitentes, como a la estructuración de la información contenida en el texto de los informes radiológicos, en el presente trabajo se abordan las tareas la automatización de la codificación clínica y la generación de conclusiones de informes de exploraciones radiológicas de próstata escritos en español. La tarea de codificación se centra en la clasificación PI-RADS v.2.1 que implementa una escala de 5 puntos basada en la probabilidad de que una combinación de hallazgos esté en correlación con la presencia de un cáncer clínicamente significativo. Con el fin de asignar estos códigos se realizaron 3 experimentos, de los cuales el mejor resultado (0,9372 de macro F1) ha mostrado un sistema basado en el ajuste fino de un modelo pre-entrenado sobre una combinación de textos médicos y clínicos. Para la tarea de generación automática de conclusiones de informes radiológicos se realizaron experimentos con un modelo de arquitectura codificadordecodificador, resultando su ajuste fino la opción preferente que proporcionó resultados prometedores (0,7545 de ROUGE-L).
Nowadays, radiology is one of the fundamental pillars of medicine by supporting both diagnosis and treatment. The radiological report, in turn, is an essential component of the study on which the interpretation of the findings by the referring physician is based. The introduction of electronic health records and the digitization of information accumulated by the health system have created a need for structuring the information contained in free text format in clinical reports to enable their use in clinical practice and medical research. In order to contribute both to the quality of communication between radiologists and referring physicians and to the structuring of the information contained in the text of the radiological reports, the present paper addresses the tasks of automation of clinical coding and the generation of conclusions of reports of radiological examinations of prostate written in Spanish. The coding task focuses on the PI-RADS v.2.1 classification that implements a 5-point scale based on the probability that a combination of findings is correlated with the presence of a clinically significant cancer. In order to assign these codes 3 experiments were performed, of which the best result (0.9372 macro F1) has been reached by a system based on fine-tuning of a pre-trained model on a combination of medical and clinical corpora. For the task of automatic generation of radiological report conclusions, experiments were carried out with an encoder-decoder architecture model, resulting its fine-tuning to be the preferred option that yields promising results (0.7545 ROUGE-L).
Nowadays, radiology is one of the fundamental pillars of medicine by supporting both diagnosis and treatment. The radiological report, in turn, is an essential component of the study on which the interpretation of the findings by the referring physician is based. The introduction of electronic health records and the digitization of information accumulated by the health system have created a need for structuring the information contained in free text format in clinical reports to enable their use in clinical practice and medical research. In order to contribute both to the quality of communication between radiologists and referring physicians and to the structuring of the information contained in the text of the radiological reports, the present paper addresses the tasks of automation of clinical coding and the generation of conclusions of reports of radiological examinations of prostate written in Spanish. The coding task focuses on the PI-RADS v.2.1 classification that implements a 5-point scale based on the probability that a combination of findings is correlated with the presence of a clinically significant cancer. In order to assign these codes 3 experiments were performed, of which the best result (0.9372 macro F1) has been reached by a system based on fine-tuning of a pre-trained model on a combination of medical and clinical corpora. For the task of automatic generation of radiological report conclusions, experiments were carried out with an encoder-decoder architecture model, resulting its fine-tuning to be the preferred option that yields promising results (0.7545 ROUGE-L).
Descripción
Categorías UNESCO
Palabras clave
Citación
Centro
Facultades y escuelas::E.T.S. de Ingeniería Informática
Departamento
Lenguajes y Sistemas Informáticos