Objective: False-positive nodule reduction is a crucial part of a computer-aided detection (CADe) system, which assists radiologists in accurate lung nodule detection. In this research, a novel scheme using multi-level 3D DenseNet framework is proposed to implement false-positive nodule reduction task. Methods: Multi-level 3D DenseNet models were extended to differentiate lung nodules from falsepositive nodules. First, different models were fed with 3D cubes with different sizes for encoding multi-level contextual information to meet the challenges of the large variations of lung nodules. In addition, image rotation and flipping were utilized to upsample positive samples which consisted of a positive sample set. Furthermore, the 3D DenseNets were designed to keep low-level information of nodules, as densely connected structures in DenseNet can reuse features of lung nodules and then boost feature propagation. Finally, the optimal weighted linear combination of all model scores obtained the best classification result in this research. Results: The proposed method was evaluated with LUNA16 dataset which contained 888 thin-slice CT scans. The performance was validated via 10-fold cross-validation. Both the Free-response Receiver Operating Characteristic (FROC) curve and the Competition Performance Metric (CPM) score show that the proposed scheme can achieve a satisfactory detection performance in the falsepositive reduction track of the LUNA16 challenge. Conclusion: The result shows that the proposed scheme can be significant for false-positive nodule reduction task.

In order to effectively aid the blind with optimal low-resolution vision and visual recovery training, pathfinding and recognition tests were performed using a simulated visual prosthetic scene. Simple and complex virtual scenes were built using 3DMAX and Unity, and pixelated to three different resolutions (32×32, 64×64, 128×128) for real-time pixel processing. Twenty subjects were recruited to complete the pathfinding and object recognition tasks within the scene. The recognition accuracy and time required were recorded and analyzed after the trials. In the simple simulated prosthetic vision scene, when the resolution was increased from 32×32 to 48×48, the object recognition time decreased from 92.19±6.97 s to 43.05±6.08 s, and the recognition accuracy increased from 51.22±8.53% to 85.52±4.93%. Furthermore, the number of collisions decreased from 10.00±2.31 to 3.00±0.68. When the resolution was increased from 48×48 to 64×64, the object recognition time further decreased from 43.05±6.08 s to 19.46±3.71 s, the recognition accuracy increased from 85.52±4.93% to 96.89±2.06%, and the number of collisions decreased from 3.00±0.68 to 1.00±0.29. In complex scenes, the time required to recognize the room type decreased from 115.00±23.02 s to 68.25±17.23 s, and object recognition accuracy increased from 65.69±9.61% to 80.42±7.70% when the resolution increased from 48×48 to 64×64. When the resolution increased from 64×64 to 128×128, the time required to recognize the room type decreased from 68.25±17.23 s to 44.88±9.94 s, and object recognition accuracy increased from 80.42±7.71% to 85.69±7.39%. Therefore, one can conclude that there are correlations between pathfinding and recognition. When the resolution increased, the time required for recognition decreased, the recognition accuracy increased, and the number of collisions decreased. Although the subjects could partially complete the recognition task at a resolution of 32×32, the recognition time was too long and recognition accuracy was not good enough to identify simple scenes. Complex scenes required a resolution of at least 48×48 for complete recognition. In addition, increasing the resolution shortened the time required to identify the type of room, and improved the recognition accuracy.

Face information is important information for identifying people, but at low resolution, face information is not well recognized. A psychological physics experiment of person recognition in daily life was designed to discover the best strategy for the limited number of stimulating electrodes, and provide useful visual information. The real-time image-processing strategy based on FaceNet were used to optimize the person information by turning complex face information into simple Chinese character information. Noted that all processed target faces were obscured by separate Chinese characters which are the target people’s last names. The psychological results showed that the image-processing strategy based on FaceNet improved recognition accuracy. The proposed strategy, to convert complex face information into simple Chinese character information, could help subjects to use their own first knowledge to identify the person who need to be identified more faster and accurately. CCS Concepts •Human-centered computing ➝ Human computer interaction; •Computing methodologies ➝ Artificial intelligence ➝ Computer vision ➝ Computer vision tasks; •Social and professional topics