Scholarlink

A malignant tumour’s chance of survival is not promising. In these situations, predicting survival becomes crucial for treating the patient. We can use the features in the segmented tumour image to predict survival. On the other hand, manually estimating survival from these images could not provide a precise survival prediction and need for automatic survival prediction arises. We have used 2D U-Net to yield segmented images with a dice coefficient of 88.1 for testing and 93.76 for training. Our proposed survival prediction workflow consists of SE-CNN which combines a convolution, max pooling, and squeeze and excitation layer to extract features that are then integrated with clinical data like age and survival days. Cox proportionality regression analysis is used to choose features with p-index less than 0.05. Ultimately, the gradient boosting algorithm is trained using these features to predict survival that gives us RMSE 264.42 which is better than the state-of-the-art methods.

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

A Robust Technique for Survival Prediction on Pre-operative MRI Images Using SE-CNN and Cox Regression

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Urvashi Dhand,
  • Najme Zehra Naqvi

摘要

A malignant tumour’s chance of survival is not promising. In these situations, predicting survival becomes crucial for treating the patient. We can use the features in the segmented tumour image to predict survival. On the other hand, manually estimating survival from these images could not provide a precise survival prediction and need for automatic survival prediction arises. We have used 2D U-Net to yield segmented images with a dice coefficient of 88.1 for testing and 93.76 for training. Our proposed survival prediction workflow consists of SE-CNN which combines a convolution, max pooling, and squeeze and excitation layer to extract features that are then integrated with clinical data like age and survival days. Cox proportionality regression analysis is used to choose features with p-index less than 0.05. Ultimately, the gradient boosting algorithm is trained using these features to predict survival that gives us RMSE 264.42 which is better than the state-of-the-art methods.