Current Issue Cover

秦俊1, 卢婷岚1, 纪柏2, 李雨晴1(1.长春理工大学;2.吉林大学白求恩第一医院)

摘 要
目的 锥形束计算机断层扫描(Cone Beam Computer Tomography, CBCT)已成为口腔诊疗领域中最常用的一项医学影像技术。由于CBCT图像本身对比度低且牙齿形状复杂,在进行牙齿分割时容易导致分割边界模糊、牙齿根部错误分割的问题。现有的方法往往无法达到预期效果,并且基于深度学习的分割网络在分割精度等性能提升到一定程度后存在生梯度爆炸、过拟合以及无法关注图像全局信息等限制。然而,牙齿分割在医生制定诊断和治疗计划方面至关重要。为了应对这一问题,提出了一种名为MF-CA Net的牙齿分割模型,以提高牙齿分割的准确性和鲁棒性。方法 MF-CA Net模型引入了多尺度特征提取模块(Multi-scale Feature Extraction Module, MFEM)和CA(Coordinate Attention)注意力机制,这些模块使网络能够准确地捕捉感兴趣的牙齿区域,并提取丰富而密集的多尺度特征信息,从而有效地指导分割任务。特别是在牙根分割方面,这些模块能够显著提高分割的精度。为了进一步增强分割算法的性能,还引入了混合损失函数,该损失函数综合考虑了像素级、局部级和全局级三个方向的牙齿边缘分割,以提高算法的准确性和稳健性。结果 实验在数据集上对MF-CA Net模型与六种主流方法进行了比较。实验结果表明,相较于其他分割方法,MF-CA Net模型在各项评价指标上都取得了显著的改进。尽管在Accuracy指标上稍低于DeeplabV3+,但在Dice评价指标上达到了0.9495的高分数,相比PyConvU-Net提高了4%,相对于DeeplabV3+提高了约4%,对比UNet提高了约16%。此外,mIoU指标提升了接近3%到11%,F2值提升了5%。结论 本文所提出的MF-CA Net网络模型可以实现对牙齿的精确分割。
Tooth segmentation network for low-dose CT

(The First Bethune Hospital of Jilin University)

Objective With the mutual penetration and integration of computer technology and modern dentistry, dentistry can be better developed and improved. Cone beam computed tomography (CBCT) has become one of the most commonly used medical imaging techniques in the field of dental diagnosis and treatment. CBCT has the advantages of low radiation dosage, simple operation, and low cost, but at the same time, the noise and artifacts of CBCT are more intense than those of conventional CT, and the fuzzy tooth boundaries will affect the doctor""s diagnosis and subsequent treatment. In oral diagnosis and treatment, doctors usually need to manually segment the tooth model in CBCT in order to formulate subsequent treatment plans, but this is not only time-consuming and labor-intensive, but also the segmentation results of the teeth are greatly affected by the subjective factors of the doctor, and the existing network method often fails to achieve the expected results, and the segmentation network based on deep learning exists in the segmentation accuracy and other performance to a certain extent after the segmentation of the network to improve the gradient explosion, over-fitting, and over-expression. limitations include gradient explosion, overfitting, and the inability to focus on global image information. Therefore, people have been working to find a dental segmentation method with high automation and high accuracy. To address this problem, a dental segmentation model called MF-CA Net is proposed, which employs a series of innovative methods to improve the accuracy and robustness of dental segmentation. Method The MF-CA Net network uses the Multi-scale Feature Extraction Module (MFEM) to extract features at different scales of the image and employs the CA (Coordinate Attention) attention mechanism that is currently excelling in improving network performance. The MFEM uses four different convolution kernels for convolution, which allows the extraction of multi-scale features and enables the network to learn more robust representations, while the dilation convolution uses four dilation rates to further increase the receptive field, which allows the network to capture more detailed information and refine important features. The CA attention mechanism calculates the spatial and channel attention weights in the input feature maps to adaptively weight them to focus on more representative features, thus focusing on the features that are more representative of the image. adaptive weighting so as to focus on the more representative local structure and global contextual information, and embedding positional information in the channel attention to help the network more accurately localize and identify objects of interest. These modules enable the network to accurately capture the tooth region of interest and extract rich and dense multi-scale feature information to effectively guide the segmentation task. Particularly for tooth root segmentation, these modules can significantly improve the accuracy of segmentation. In addition, in order to further improve the performance of the segmentation algorithm, the MF-CA Net network model also employs structural similarity to construct the boundary loss function, and the algorithm uses a mixture of the Dice loss function, the binary cross-entropy loss function, and the SSIM loss function as the final loss function; the Dice loss function is used to compute the similarity between the two sets of images, and the cross-entropy loss function is used to predict the segmentation result and the pixels corresponding to the real segmentation result. This loss function integrates the tooth edge segmentation in three directions: pixel level, local level, and global level to improve the accuracy and robustness of the algorithm. Result To more accurately evaluate the performance of the proposed model in the tooth segmentation task, Dice Similarity Coefficient (DSC), Mean Intersection to Merger Ratio (mIoU), Accuracy, Recall, Precision, and F2 Score are used as evaluation metrics. This paper compares the MF-CA Net model and six mainstream methods on the dataset. The experimental results show that the MF-CA Net model has significant improvement in most of the evaluation metrics compared with other segmentation methods. Although it is slightly lower than DeeplabV3+ in accuracy metrics, it achieves a high score of 0.9495 in Dice evaluation metrics, which is an improvement of 4% compared with PyConvU-Net, about 4% compared with DeeplabV3+, and about 16% compared with UNet. In addition, the mIoU metric improves from 3% to nearly 11%; the precision value reaches 0.9421, which is a 7% improvement compared to UNET++; the recall metric reaches 0.9687, which is an 8% improvement compared to the UNET network; and the F2 metric reaches 0.9543, which is a 5% improvement compared to the Res-UNet value. These results fully demonstrate the superiority of the MF-CA network model in the tooth segmentation task. Conclusion In summary, the MF-CA network model proposed in this paper successfully solves the difficult problem of tooth segmentation in CBCT images by introducing a multiscale feature extraction module, an attention mechanism, and a hybrid loss function. Many experimental results verify the model""s superiority in accurate tooth segmentation, and the model is expected to be widely used in dental diagnosis and treatment, which is of great significance in oral diagnosis and treatment.