ODR3DNet: Omni-Dimension Dynamic Residual 3D Net for Pulmonary Nodule Detection

Lung cancer remains a potentially fatal global health concern, underscoring the criticality of enhancing the precision of early lung cancer diagnosis to ameliorate patient prognoses. Over the past years, the ascent of deep learning (DL) has ushered in a formidable arsenal, with DL-based computer-aided systems orchestrating remarkable strides in the realm of pulmonary nodule detection (PND). This investigation introduces a pioneering approach, namely the Omni-dimension Dynamic Residual 3D Net (ODR3DNet), meticulously tailored for PND by harnessing the prowess of full-dimensional dynamic 3D convolution. The core thrust behind ODR3DNet lies in its ability to surmount the constraints bedeviling conventional 3D CNNs. These limitations encompass a lack of flexibility and a constrained feature extraction capacity. When juxtaposed with the conventional PND algorithms that dominate the field, our proposed algorithm unfurls its prowess by securing an impressively high CPM score of 0.885, thus establishing its resounding superiority. What’s more, our exploration delves deeper as ablation experiments are harnessed to substantiate the manifold contributions of OD3D towards bolstering performance, all while offering an optimal configuration for seamless integration.