Fig. 5 | Scientific Reports

Fig. 5

From: Muscle-Driven prognostication in gastric cancer: A multicenter deep learning framework integrating Iliopsoas and erector spinae radiomics for 5-Year survival prediction

Fig. 5

Performance comparison of 3D neural networks using ROC curves. The evaluation of DenseNet121 and DenseNet201 models on abdominal CT scans showed limited classification capability, with AUC values below 0.5 for both models. While 3D architectures can capture spatial information effectively, these results suggest they may not be optimal for this particular task. The ROC curves (left: DenseNet121; right: DenseNet201) with 95% confidence intervals demonstrate this performance limitation, leading to their exclusion from the final model integration.

Back to article page