Application of radiomics-based prediction model to predict preoperative lymph node metastasis in prostate cancer: a
Yanghuang Zheng1, Yuelin Du1, Biao Zhang1
1Department of Urology, Lanzhou University Second Hospital, Lanzhou, Gansu, China.
Background:
This study aims to comprehensively evaluate the accuracy and efficacy of radiomics models based on imaging equipment in predicting prostate cancer (PCa) lymph node metastasis (LNM).
Methods:
We systematically searched PubMed, Embase, Cochrane Library, Web of Science, and Sinomed databases from their establishment until July 2024. The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) criteria and the Radiomics Quality Score (RQS) tools were utilized to assess the quality of the studies. Indicators such as the pooled area under the curve (AUC), sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio were computed to evaluate the predictive effect of radiomics technology on LNM of PCa.
Results:
A total of 1860 patients diagnosed with LNM of PCa through histological examination were included in this meta-analysis. The radiomics model for predicting LNM in PCa showed a pooled AUC value of 0.88 (95% confidence interval (CI) [0.85 - 0.91]), with a sensitivity and specificity of 0.81 (95% CI [0.64 - 0.91]) and 0.85 (95% CI [0.75 - 0.91]), respectively. The positive likelihood ratio was 5.43 (95% CI [3.34 - 8.84]), the negative likelihood ratio was 0.22 (95% CI [0.12 - 0.43]), and the diagnostic odds ratio was 24.21 (95% CI [10.59 - 55.32]). The meta-analysis showed significant heterogeneity among the included studies. No threshold effect was detected. The subgroup analysis showed that the least absolute shrinkage and selection operator regression algorithm had the higher diagnostic sensitivity, with a pooled sensitivity of 0.96 (95% CI [0.90 - 1.00]) (p = 0.02), while the random forest algorithm was the opposite, with a pooled sensitivity of 0.48 (95% CI [0.16 - 0.80]) (p = 0.01). Radiomics features without intraclass correlation coefficient preprocessing would lead to a decrease in diagnostic specificity, 0.73 (95% CI [0.53 - 0.92]) (p = 0.04). The pooled specificity with an RQS score≥ 17 was 0.77 (95% CI [0.65 - 0.88]) (p = 0.01), and the higher the score, the lower the diagnostic specificity would be.
Conclusions:
The predictive model based on radiomics features has the potential to serve as an auxiliary approach for predicting preoperative LNM of PCa.
Systematic Review Registration:
https://www.crd.york.ac.uk/prospero/, identifier PROSPERO CRD42024575818.
