Estimação de volume mamário em cirurgias plásticas do Sistema Único de Saúde, utilizando aprendizado de máquina.
Introduction: Breast volume estimation is a crucial step in mammoplasty, directly influencing the accuracy of results and the efficiency of surgical time. Despite its importance, there is no practical and objective methodology available for this task, and it is often dependent on the subjective perc...
| Autor: | |
|---|---|
| Formato: | tesis de maestría |
| Estado: | Versión publicada |
| Fecha de publicación: | 2024 |
| País: | Brasil |
| Recursos: | Universidade Federal do Maranhão (UFMA) |
| Repositorio: | Biblioteca Digital de Teses e Dissertações da UFMA |
| Idioma: | portugués |
| OAI Identifier: | oai:tede2:tede/6131 |
| Acesso em linha: | https://tedebc.ufma.br/jspui/handle/tede/6131 |
| Access Level: | acceso abierto |
| Palavra-chave: | mamoplastia; mama; cirurgia plástica; aprendizado de máquina. mammoplasty; breast; plastic surgery; machine learning. Saúde Coletiva |
| Resumo: | Introduction: Breast volume estimation is a crucial step in mammoplasty, directly influencing the accuracy of results and the efficiency of surgical time. Despite its importance, there is no practical and objective methodology available for this task, and it is often dependent on the subjective perception of the surgeon. Therefore, this study aimed to propose an innovative technique for breast volume estimation, based on 2D digital photographs and Machine Learning. Methods: Twenty-five women with symptomatic breast hypertrophy were invited, whose breasts were photographed in a standardized manner, in hospital environment. A Machine Learning technique capable of localizing, segmenting and estimating breast volume from images was developed. Results: For the breast segmentation task, the U-Net network provided excellent performance, reaching Dice coefficients of 0.97. Regression algorithms were used to improve the volume estimates of the new technique, which proved to be more accurate and consistent than those performed subjectively by invited surgeons. The mean absolute error (MAE) of the new methodology was 206.722 mL compared to magnetic resonance imaging (gold standard), with the SVM (support vector machines) model having the best performance. Bland-Altmann plots confirmed the reliability of the proposed method, compared to current clinical practice. Conclusion: This work paves the way for a new approach for simple, accessible, practical and objective breast volume estimation, based on 2D digital photographs and Machine Learning, which can redefine current standards. |
|---|