Please click here to download the model.
Abstract:
Pancreatic ductal adenocarcinoma (PDAC) is the fourth most common cause of global cancer mortality. It is imperative to develop a prognostic tool to identify high-risk with lymph node metastasis(LNM) patients and assist in advance.
The data of 148 patients (75 females and 73 males; mean age, 59.2 ± 10.6 years) with 150 PDAC lesions (mean diameter, 2.9 ± 0.9 cm) at our cancer center were screened for eligibility. 113 patients were assigned to the training set (from August 5, 2016 to 31, December, 2019), 35 were assigned to the internal validation set (from 1, January 2020 to 11, October, 2020).
To investigate the optimal DLR model in predicting LNM, four DLR models based on DECT image (100 keV, 150 keV, VMI 40 keV and 100-150 keV) were compared, C-DLR model incorporating DECT DLR signature and clinical variables including CT-reported T stage, LN status, glutamyl transpeptadase and glucose improved significantly performance in predicting LNM (P < 0.05).
We added a linear layer to the fully connected layer (FC) in the DLR model to reduce the output features from 512 to 64. The features extracted from the DLR model are input into the SVM, where the SVM replaces the last layer of the fully connected layer and plays the role of output probability. We provide the parameters of the DLR model, the features extracted from the DLR, and the related code of the SVM. You can test your data using our model code and loading the model weights (PyTorch> 1.1.0).
The detailed information, please refer to our paper (under review).
