Kabak S L, Karapetyan G M, Melnichenko Y M, Savrasova N A, Kosik I I
Belarusian State Medical University, Minsk, Belarus.
LLC «LODE», Minsk, Belarus.
Vestn Otorinolaringol. 2021;86(2):49-53. doi: 10.17116/otorino20218602149.
Was to compare manual, semi-automatic and automatic methods for determining the maxillary sinus volume using cone beam computed tomography (CBCT).
CBCT images from 48 patients (96 maxillary sinuses) with no history of sinus and alveolar bone surgery, who were presented to Minsk medical centers, were used in this study. Neural network training was performed on CBCT scans of 42 patients (84 maxillary sinuses).The height, depth and width of the sinus were measured manually on CBCT scans of 6 patients (12 maxillary sinuses). Maxillary sinus volume (V) was calculated by the formula: V=height´depth´1/3 width. Semi-automatic segmentation was carried out by an expert radiologist. The convolutional neural network technology was applied for maxillary sinus automatic segmentation.
The largest values were revealed by using the automatic method for sinus volume measurement. These values were within the 95% confidence interval (±4.29 cm) of the average sinus volume obtained from semi-automatic method.
The data obtained using the convolutional neural network technique (artificial intelligence) has a high correlation with the results of sinus morphometric analysis acquired through manual and semi-automatic methods. Automatic maxillary sinus segmentation technique does not require special user knowledge. This method is reproducible and it is implemented in a short time interval.
比较使用锥形束计算机断层扫描（CBCT）测定上颌窦容积的手动、半自动和自动方法。
本研究使用了明斯克医疗中心48例（96个上颌窦）无鼻窦及牙槽骨手术史患者的CBCT图像。对42例患者（84个上颌窦）的CBCT扫描进行神经网络训练。在6例患者（12个上颌窦）的CBCT扫描上手动测量鼻窦的高度、深度和宽度。上颌窦容积（V）通过公式计算：V = 高度×深度×1/3宽度。由专业放射科医生进行半自动分割。应用卷积神经网络技术进行上颌窦自动分割。
使用自动方法测量鼻窦容积得到的值最大。这些值在半自动方法获得的平均鼻窦容积的95%置信区间（±4.29 cm）内。
使用卷积神经网络技术（人工智能）获得的数据与通过手动和半自动方法获得的鼻窦形态测量分析结果具有高度相关性。上颌窦自动分割技术不需要用户具备特殊知识。该方法具有可重复性，且能在短时间内实现。
