Scholarlink

Purpose <p>To compare whole-lesion apparent diffusion coefficient (ADC) histogram analysis with representative ADC value for assessing muscle invasion in bladder cancer, and to build a decision tree model that combines bi-parametric Vesical Imaging Reporting and Data System (bp VI-RADS) with ADC values.</p> Methods <p>This retrospective study included 82 patients with bladder cancer who underwent 3T MRI and transurethral resection. The bp VI-RADS was scored using T2-weighted and diffusion-weighted imaging, and ADC maps. For histogram analysis, tumors were segmented on all ADC slices to obtain ADC percentiles. For a representative ADC, three regions of interest were placed in visually lowest ADC areas and averaged to obtain the minimum mean ADC. ADC parameters were compared between muscle invasive (MIBC) and non-muscle invasive (NMIBC) disease. Diagnostic performance was assessed using receiver operating characteristic curve analysis with area under the curve (AUC), and combination models were constructed using logistic regression and decision tree analysis.</p> Results <p>Among histogram-derived parameters, the 25th percentile ADC achieved the highest AUC, and the minimum mean ADC was strongly correlated with it. As single parameters, bp VI-RADS, the 25th percentile and the minimum mean ADC showed similar accuracy (0.74–0.76). Logistic regression models combining bp VI-RADS with the 25th percentile or the minimum mean ADC achieved higher accuracy (0.88 and 0.87). Decision tree models using bp VI-RADS and ADC reached accuracy of 0.82 and 0.80. In both decision tree models, ADC provided additional stratification primarily for bp VI-RADS 4 lesions, using cutoffs of 1,183 × 10<sup>⁻6</sup> mm²/s (25th percentile ADC) and 1,100 × 10<sup>⁻6</sup> mm²/s (minimum mean ADC).</p> Conclusion <p>The 25th percentile and the minimum mean ADC showed similar diagnostic performance for predicting muscle invasion. A decision tree combining bp VI-RADS with ADC measurements may provide an interpretable and clinically applicable approach, particularly for bp VI-RADS 4 lesions.</p>

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

Decision tree models combining Bi-parametric vesical imaging reporting and data system and apparent diffusion coefficient metrics for predicting muscle-invasive bladder cancer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月23日

查看全文
  • Daichi Sugawara,
  • Kento Hatakeyama,
  • Motoko Konno,
  • Tomoki Tozawa,
  • Tomochika Sato,
  • Toshiki Murata,
  • Emika Murasawa,
  • Shintaro Narita,
  • Hiroshi Nanjo,
  • Tomonori Habuchi,
  • Naoko Mori

摘要

Purpose

To compare whole-lesion apparent diffusion coefficient (ADC) histogram analysis with representative ADC value for assessing muscle invasion in bladder cancer, and to build a decision tree model that combines bi-parametric Vesical Imaging Reporting and Data System (bp VI-RADS) with ADC values.

Methods

This retrospective study included 82 patients with bladder cancer who underwent 3T MRI and transurethral resection. The bp VI-RADS was scored using T2-weighted and diffusion-weighted imaging, and ADC maps. For histogram analysis, tumors were segmented on all ADC slices to obtain ADC percentiles. For a representative ADC, three regions of interest were placed in visually lowest ADC areas and averaged to obtain the minimum mean ADC. ADC parameters were compared between muscle invasive (MIBC) and non-muscle invasive (NMIBC) disease. Diagnostic performance was assessed using receiver operating characteristic curve analysis with area under the curve (AUC), and combination models were constructed using logistic regression and decision tree analysis.

Results

Among histogram-derived parameters, the 25th percentile ADC achieved the highest AUC, and the minimum mean ADC was strongly correlated with it. As single parameters, bp VI-RADS, the 25th percentile and the minimum mean ADC showed similar accuracy (0.74–0.76). Logistic regression models combining bp VI-RADS with the 25th percentile or the minimum mean ADC achieved higher accuracy (0.88 and 0.87). Decision tree models using bp VI-RADS and ADC reached accuracy of 0.82 and 0.80. In both decision tree models, ADC provided additional stratification primarily for bp VI-RADS 4 lesions, using cutoffs of 1,183 × 10⁻6 mm²/s (25th percentile ADC) and 1,100 × 10⁻6 mm²/s (minimum mean ADC).

Conclusion

The 25th percentile and the minimum mean ADC showed similar diagnostic performance for predicting muscle invasion. A decision tree combining bp VI-RADS with ADC measurements may provide an interpretable and clinically applicable approach, particularly for bp VI-RADS 4 lesions.