Análise multicritério da qualidade de imagens PET/CT: uma abordagem quantitativa baseada em análise de componentes principais

Abstract

In this dissertation, a methodology is presented for modeling and analyzing medical images in the DICOM format obtained from PET/CT scans, with an emphasis on optimizing image quality with reduced radiopharmaceutical doses through Principal Component Analysis (PCA). The proposed approach enables image processing and visualization, as well as the extraction and quantitative evaluation of the principal features of the data, using metrics such as PSNR, SSIM, MSE, MAE, and entropy difference. These metrics allow for a comprehensive analysis, considering both visual similarity and the preservation of information while minimizing errors in reconstructed images. The dimensionality reduction performed by PCA proved effective in preserving the main structural and statistical characteristics of the images, resulting in a more compact and efficient data representation. The integrated use of evaluation metrics in a combined ranking highlighted quality variations across different cases, with results exceeding 0.8 in several patients, demonstrating high similarity in preserving the characteristics of the original images. Additionally, the inclusion of entropy difference provided a complementary perspective to traditional metrics, assessing changes in the degree of disorder or complexity—an important aspect for clinical interpretation. As the main contributions, this dissertation presents an approach that integrates PCA and radiopharmaceutical dose reduction in PET/CT images based on dimensionality reduction, the application of normalized image evaluation metrics with a proposed combined ranking using individual weights, and the estimation of metrics for comparison between computational analyses and professional assessments. The results underscore the relevance of an integrated quantitative approach for medical image analysis, highlighting the potential of PCA as a tool for preprocessing and dimensionality reduction in PET/CT. Future studies may explore the optimization of metric weights in the combined ranking and extend the application of the methodology to other clinical contexts and imaging modalities, further establishing its significance in biological information processing.