Scholarlink

We explored the performance potential of radiomics analysis of PSMA-PET images for predicting response to PSMA-targeted therapies in high-grade prostate cancer (PCa). Radiomics analysis was performed on 18F-PSMA-1007 PET images of 18 patients with histologically confirmed or verified disease at 2-year follow-up. The radiomics workflow involved segmenting regions of interest (ROIs) from images, extracting features from ROIs, selecting features using a hybrid descriptive-inferential approach. A hybrid descriptive-inferential approach combines data summarization with statistical testing. Descriptive statistics characterize the dataset (e.g., mean, SD), while inferential methods assess significance and generalizability. This dual approach provides both insight and evidence-based conclusions. Among the 112 radiomics features, original_gray level co-occurrence matrix (glcm)_DifferenceEntropy and original_first-order_Entropy performed exemplary in predicting E-PSMA scores from primary PCa lesions with AUROC 100%, \(p < 0.001\) . The original_Gray Level Run Length Matrix (glrlm)_LongRunLowGrayLevelEmphasis feature showed suboptimal performance in providing PRIMARY scores with AUROC 70.8%, \(p = 0.02\) . Radiomic analysis of PSMA PET images has clinical application in predicting internationally validated PSMA-PET scoring systems in PCa staging, establishing a roadmap for increased efficiency in predicting PRIMARY score in TM PCa staging.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Preliminary Study Exploring the Potential of ML Models Applied to Radiomic Features Extracted from PSMA-PET: Prediction of the PRIMARY and E-PSMA Scores in the Evaluation of Primary PCa Lesions

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Viviana Benfante,
  • Pierpaolo Purpura,
  • Costanza Longo,
  • Albert Comelli,
  • Pierpaolo Alongi

摘要

We explored the performance potential of radiomics analysis of PSMA-PET images for predicting response to PSMA-targeted therapies in high-grade prostate cancer (PCa). Radiomics analysis was performed on 18F-PSMA-1007 PET images of 18 patients with histologically confirmed or verified disease at 2-year follow-up. The radiomics workflow involved segmenting regions of interest (ROIs) from images, extracting features from ROIs, selecting features using a hybrid descriptive-inferential approach. A hybrid descriptive-inferential approach combines data summarization with statistical testing. Descriptive statistics characterize the dataset (e.g., mean, SD), while inferential methods assess significance and generalizability. This dual approach provides both insight and evidence-based conclusions. Among the 112 radiomics features, original_gray level co-occurrence matrix (glcm)_DifferenceEntropy and original_first-order_Entropy performed exemplary in predicting E-PSMA scores from primary PCa lesions with AUROC 100%, \(p < 0.001\) . The original_Gray Level Run Length Matrix (glrlm)_LongRunLowGrayLevelEmphasis feature showed suboptimal performance in providing PRIMARY scores with AUROC 70.8%, \(p = 0.02\) . Radiomic analysis of PSMA PET images has clinical application in predicting internationally validated PSMA-PET scoring systems in PCa staging, establishing a roadmap for increased efficiency in predicting PRIMARY score in TM PCa staging.