Chi Xiao

Segmentation and analysis of microscopic optical images is a fundamental task in the field of biomedicine. However, efficiently, accurately, and robustly segmenting regions of interest in these images presents a significant challenge. The segment anything model (SAM) has demonstrated remarkable generalization capabilities in natural image segmentation tasks, revealing its potential for segmenting microscopic optical images. In this study, we propose Brain-SAM, a general automatic segmentation model based on SAM for the automatic segmentation of microscopic optical images. Specifically, we introduce an automatic prompt encoder to enable high-throughput automated segmentation of these images. Additionally, we propose a segmentation optimizer to further enhance the model’s segmentation performance. Testing on eight benchmark datasets, representing common scenarios in microscopic optical image segmentation, shows that Brain-SAM outperforms specialized segmentation models in the vast majority of segmentation tasks. Notably, on the Brain, Tek and Lectin3d datasets, Brain-SAM achieved IoU scores of 98.07%, 93.13% and 88.49% respectively, along with Dice scores of 99.03%, 96.44% and 93.89%. Moreover, we provide a series of rich, publicly available brain science image datasets created using fluorescence microscopic optical tomography (fMOST) technology.

The brain is the least explored organ in the human body. Brain functions are realized through a complex neural network composed of a vast number of neurons, and understanding the morphology of these neurons is base to brain studies. However, obtaining high-quality, large-scale data on neuron morphology remains a significant challenge. In this study, we propose a precise data-graded allocation method for neuron reconstruction, the accuracy is safeguarded by the allocation algorithm and the quantitative model. Reconstruction efficiency was improved by optimizing automated reconstruction algorithm, human-machine interaction workflow and human-task matching method. We have implemented this strategy on a web-based platform, and the results show that 92.9% of image data can be easily reconstructed, thereby reducing the skill requirements for participant. The reconstruction accuracy is 98.2%$\pm$3.1%, better than existing methods. We also provides meticulously annotated datasets that can propel significant advancements in artificial intelligence technology. In addition, we can offer a well-balance across quality, cost, and efficiency, sharing a more flexible and versatile solution for three-dimensional neuron reconstruction.

The rapid development of optical labeling and imaging technologies has enabled scientists to capture three-dimensional images of mammalian brains at the single-neuron level. However, it has also brought about numerous technical challenges. In neuronal images, fibers exhibit lower fluorescence intensity than cell bodies and are densely packed, making it difficult to distinguish fiber signals from background noise or resolve connectivity accurately. While sparse high-brightness labeling and sensitive imaging technologies have partially addressed these issues, they have not been fundamentally resolved. This study focused on data post-processing and proposed an image enhancement method using a fiber topology-preserving multi-module fusion network. By incorporating a self-attention mechanism and a topology-preserving loss function, the method enhanced the signal-to-noise ratio and continuity of neuronal fibers. Applied to the three-dimensional reconstruction of complex fiber structures, it significantly improved the performance of existing algorithms, offering an effective technical approach for precise neuronal fiber structure analysis.

Citrus rind color is a good indicator of fruit development, and methods to monitor and predict color transformation therefore help the decisions of crop management practices and harvest schedules. This work presents the complete workflow to predict and visualize citrus color transformation in the orchard featuring high accuracy and fidelity. A total of 107 sample Navel oranges were observed during the color transformation period, resulting in a dataset containing 7,535 citrus images. A framework is proposed that integrates visual saliency into deep learning, and it consists of a segmentation network, a deep mask-guided generative network, and a loss network with manually designed loss functions. Moreover, the fusion of image features and temporal information enables one single model to predict the rind color at different time intervals, thus effectively shrinking the number of model parameters. The semantic segmentation network of the framework achieves the mean intersection over a union score of 0.9694, and the generative network obtains a peak signal-to-noise ratio of 30.01 and a mean local style loss score of 2.710, which indicate both high quality and similarity of the generated images and are also consistent with human perception. To ease the applications in the real world, the model is ported to an Android-based application for mobile devices. The methods can be readily expanded to other fruit crops with a color transformation period. The dataset and the source code are publicly available at GitHub.

Gradient Learning (GL), aiming to estimate the gradient of target function, has attracted much attention in variable selection problems due to its mild structure requirements and wide applicability. Despite rapid progress, the majority of the existing GL works are based on the empirical risk minimization (ERM) principle, which may face the degraded performance under complex data environment, e.g., non-Gaussian noise. To alleviate this sensitiveness, we propose a new GL model with the help of the tilted ERM criterion, and establish its theoretical support from the function approximation viewpoint. Specifically, the operator approximation technique plays the crucial role in our analysis. To solve the proposed learning objective, a gradient descent method is proposed, and the convergence analysis is provided. Finally, simulated experimental results validate the effectiveness of our approach when the input variables are correlated.