Central Feature Network Enables Accurate Detection of Both Small and Large Particles in Cryo-Electron Tomography

The advancements in Cryo-Electron Tomography (cryo-ET) have made it possible to visualize molecules in their natural cellular settings in three-dimensional space. Such visualizations play a crucial role in investigating the functions of biological entities under native conditions. Recently, deep learning techniques have proven effective in addressing the challenge of detecting particles in cryo-ET data. Nevertheless, the task of precisely identifying and categorizing multi-class molecules remains difficult due to factors such as the low signal-to-noise ratio and the diverse range of sizes in particle selection. In this study, we present a new framework called Central Feature Network (CFN) for detecting objects in 3D and implement it in cryo-ET analysis. A key strength of CFN is its ability to integrate central features across different scales, enabling the accurate detection of both small and large molecules. In comparison to existing methods, CFN enhances the F1 score for classification by \(3.6\%\) , \(7.3\%\) , \(6.6\%\) , and \(5.1\%\) for the four smallest molecules tested, while maintaining similar or higher F1 scores for other molecules examined. Our code is available at https://github.com/Wangyaoyuu/cfn_scr .