Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed framework uses an attention-based Multiple Instance Learning (MIL) model with a ConvNeXt backbone for lung disease diagnosis from chest CT volumes.method 95%
• A Gradient Reversal Layer (GRL) is integrated to adversarially suppress gender-predictive information in the learned scan representation.method 95%
• The model achieves a mean validation competition score of 0.685 with a standard deviation of 0.030.result 95%
• The best single fold of the model achieved a score of 0.759.result 95%
• The training process incorporates focal loss with label smoothing and stratified cross-validation.method 90%
• Ensembling of five-fold checkpoints with horizontal-flip test-time augmentation is used for inference.method 90%
• The framework addresses the challenge of sparse pathological signals across hundreds of CT slices.technical 90%
• The framework is designed to mitigate severe demographic imbalance across disease class and gender.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card