(Thursday, 7/16/2020) 10:30 AM - 12:30 PM [Eastern Time (GMT-4)]
Room: Track 4
Purpose: lymph node (LN) malignancy prediction is essential to designing radiation therapy treatment plans for head and neck (HN) cancer patients. To predict LN malignancy status accurately, we proposed an attention-guided-classification (AGC) scheme that doesn’t require accurate delineation of LNs, but can highlight the discriminative region which is important for determining malignancy status by the model itself.
Methods: the proposed AGC scheme, there is an attention-guided-CNN (agCNN) module followed by a classification-CNN (cCNN) module. The input of the proposed AGC scheme is a region-of-interest (ROI) patch containing the LN and its surrounding tissues. The agCNN module is designed to find the discriminative region in the ROI, which outputs an activation map whose voxel values indicate the importance of the voxels in the final malignancy prediction. Through multiplying the activation map with the ROI, we obtain the input for the cCNN module which finally outputs the LN malignancy probability. The training objective function for the proposed AGC scheme is formulated as a combination of 1) discriminative region loss measured by the dice similarity coefficient index between the activation map and the LN contour; and 2) malignancy prediction loss measured by the binary cross entropy between the ground truth label and the predicted probabilities. This study included 791 LNs contoured on PET and contrast CT from 129 patients with HN cancer. Five folder cross validation was used to evaluate the performance of the proposed AGC scheme.
Results: sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC) values obtained by the proposed AGC scheme were 0.91, 0.93, 0.92 and 0.98 respectively.
Conclusion: achieving accurate malignancy prediction, the proposed AGC scheme can highlight the discriminative region in the ROI, providing further useful information to physicians for contouring LNs and offering the interpretability of the prediction model.
Keywords
CAD, Image Analysis, ROC Analysis
Taxonomy
IM- Dataset Analysis/Biomathematics: Machine learning
Contact Email
