DR-ConvNeXt: A Dilated Residual Network based on ConvNeXt for Lung Nodule Features Prediction

The increasing application of Low Dose Spiral CT (LDCT) has enhanced lung nodule detection, promoting early diagnosis and better cure outcomes. Addressing the challenges of false positives and the radiologists’ growing workload, we refined the ConvNeXt model by integrating dilated residual modules for precise lung nodule detection. This led to a 9.48% improvement in 2D accuracy on the Luna16 dataset, attaining 87.24%. The 3D feature accuracy reached 81.78%, a leap of 21.6%, in 10-fold cross-validation. For the LNDB dataset, 2D classification surged to 95.74% (a 1.78% boost), and 3D prediction precision escalated from 72.02% to 80.73%.