Bowen Zhao, Changkai Ji, Yuejie Zhang, Wen He, Yingwen Wang, Qing Wang, Rui Feng, Xiaobo Zhang. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Background: In solid-predominantly invasive lung adenocarcinoma (SPILAC), occult lymph node metastasis (OLNM) is pivotal for determining treatment strategies. This study seeks to develop and validate a fusion model combining radiomics and deep learning to predict OLNM preoperatively in SPILAC patients across multiple centers. Methods: In this study, 1325 cT1a-bN0M0 SPILAC patients from six hospitals were retrospectively analyzed and divided into pathological nodal positive (pN+) and negative (pN-) groups. Three predictive models for OLNM were developed: a radiomics model employing decision trees and support vector machines; a deep learning model using ResNet-18, ResNet-34, ResNet-50, DenseNet-121, and Swin Transformer, initialized randomly or pre-trained on large-scale medical data; and a fusion model integrating both approaches using addition and concatenation techniques. The model performance was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC). Results: All patients were assigned to four groups: training set (n = 470), internal validation set (n = 202), and independent test set 1 (n = 227) and 2 (n = 426). Among the 1325 patients, 478 (36%) had OLNM (pN+). The fusion model, combining radiomics with pre-trained ResNet-18 features via concatenation, outperformed others with an average AUC (aAUC) of 0.754 across validation and test sets, compared to aAUCs of 0.715 for the radiomics model and 0.676 for the deep learning model. Conclusion: The radiomics-deep learning fusion model showed promising ability to generalize in predicting OLNM from CT scans, potentially aiding personalized treatment for SPILAC patients across multiple centers.

Detecting in open-world scenarios poses a formidable challenge for models intended for real-world deployment. The advanced closed set object detectors achieve impressive performance under the closed set setting, but often produce overconfident misprediction on unknown objects due to the lack of supervision. In this paper, we propose a novel Evidential Object Detector (EOD) to formulate the Open Set Object Detection (OSOD) problem from the perspective of Evidential Deep Learning (EDL) theory, which quantifies classification uncertainty by placing the Dirichlet Prior over the categorical distribution parameters. The task-specific customized evidential framework, equipped with meticulously designed model architecture and loss function, effectively bridges the gap between EDL theory and detection tasks. Moreover, we utilize contrastive learning as an implicit means of evidential regularization and to encourage the class separation in the latent space. Alongside, we innovatively model the background uncertainty to further improve the unknown discovery ability. Extensive experiments on benchmark datasets demonstrate the outperformance of the proposed method over existing ones.