Skin lesion monitoring and reconstruction using computational methods and monte carlo simulations in biomedical optics

• Ano de defesa: 2025
• Autor(a) principal: Palamoni, Otávio Perez
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: 3D reconstruction; Análise de lesões; Estimativa de volume; Imagem médica; Lesion analysis; Medical imaging; Monte Carlo simulation; Reconstrução 3D; Simulação de Monte Carlo; Volume estimation
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/76/76135/tde-06102025-080843/

Accurate volumetric analysis of skin lesions plays a crucial role in both clinical and preclinical research. In the clinical setting, it provides objective metrics for monitoring disease progression, assessing treatment response, and guiding therapeutic decisions. In preclinical studies, reliable lesion quantification is equally important for evaluating experimental therapies and establishing ethical endpoints in animal models. Despite its importance, conventional measurement techniques—such as manual use of calipers and rulers—are limited by operator variability and reduced reproducibility, highlighting the need for more robust and automated approaches.This work presents a computational pipeline developed to estimate lesion volumes from single RGB images, without relying on dedicated depth sensors. The method combines monocular depth estimation—using the MiDaS model—with 3D reconstruction and voxelization, allowing lesion morphology to be preserved and volume to be computed through voxel counting. The pipeline was tested in stages of increasing complexity. Initially, simulations with virtual objects of known geometry were used to confirm the mathematical reliability of the approach. The method was then applied to physical models, including 3D-printed and handcrafted objects, showing strong agreement between estimated and reference volumes (above 98% in ideal conditions). In vivo validation was conducted using murine models with cutaneous lesions, demonstrating the feasibility of applying the method in more realistic and variable conditions. In addition to volume estimation, the reconstructed voxelized structures were incorporated into light transport simulations using the Monte Carlo eXtreme (MCX) framework. This integration demonstrated that depth-based reconstructions can be used to create spatial domains for photon propagation modeling, which may be useful in biomedical optics applications such as phototherapy planning or optical diagnostics. The pipeline proved to be low-cost, accessible, and adaptable. It runs on standard computers and can be used without the need for specialized equipment, making it a potentially valuable tool for laboratories or clinical research environments. Scripts and tools developed for this project are available in open-source platforms to support future improvements and broader use in research or applied settings.