Application of knowledge distillation technology for fine segmentation of three-vessel and trachea views in fetal echocardiographic images / 中华超声影像学杂志

Objective:To explore the application value of fetal heart ultrasound image segmentation network model based on knowledge distillation technology in the fine segmentation of fetal heart ultrasound image at three-vessel and trachea (3VT) views.Methods:One thousand and three hundred fetals were retrospectively collected from Sir Run Run Shaw Hospital, Zhejiang University College of Medicine from January 2016 to December 2021, the two-dimensional grayscale ultrasound images of fetal heart at 3VT views were analyzed and then divided into training, validation, and test sets. The training and validation sets were used to construct the auxiliary diagnostic network models, and the test set was used to test the reliability of different network models (U-Net, DeepLabv3+ ). The 3VT views were collected and annotated by an experienced doctor as the reference standard. The intersection over union (IoU), pixel accuracy (PA) and Dice coefficient (Dice) were used as the 3 indexes to evaluate the segmentation accuracy, and the diagnostic efficiency of the training model was evaluated. The training model and the most commonly used segmentation models were identified, and the results were compared. A total of 101 images were randomly selected and assigned to junior doctors, AI and junior doctors assisted AI interpretation. Bland-Altman images were drawn to evaluate their consistency with the reference standard, and the results were compared.Results:The training model of knowledge distillation algorithm achieved better results than U-Net, DeepLabv3+ models on all evaluation indexes, and the average IoU, PA and Dice were 68.6%, 81.4% and 81.3%, respectively. Compared with the U-Net model and DeepLabv3+ model, more accurate segmentation boundaries were obtained by the knowledge distillation algorithm training model, and the quantitative evaluation indexes were improved. With the aid of the model, the diagnostic accuracy of junior doctors was improved.Conclusions:The knowledge distillation algorithm training model segmentation method can identify the anatomical structure of the fetal heart in the 3VT view of the fetal heart ultrasound image, and the recognition result is obviously better than other related methods, and can improve the accuracy of image recognition for doctors with low experience.

Automated Classification of Inherited Retinal Diseases in Optical Coherence Tomography Images Using Few-shot Learning / 生物医学与环境科学（英文）

OBJECTIVE@#To develop a few-shot learning (FSL) approach for classifying optical coherence tomography (OCT) images in patients with inherited retinal disorders (IRDs).@*METHODS@#In this study, an FSL model based on a student-teacher learning framework was designed to classify images. 2,317 images from 189 participants were included. Of these, 1,126 images revealed IRDs, 533 were normal samples, and 658 were control samples.@*RESULTS@#The FSL model achieved a total accuracy of 0.974-0.983, total sensitivity of 0.934-0.957, total specificity of 0.984-0.990, and total F1 score of 0.935-0.957, which were superior to the total accuracy of the baseline model of 0.943-0.954, total sensitivity of 0.866-0.886, total specificity of 0.962-0.971, and total F1 score of 0.859-0.885. The performance of most subclassifications also exhibited advantages. Moreover, the FSL model had a higher area under curves (AUC) of the receiver operating characteristic (ROC) curves in most subclassifications.@*CONCLUSION@#This study demonstrates the effective use of the FSL model for the classification of OCT images from patients with IRDs, normal, and control participants with a smaller volume of data. The general principle and similar network architectures can also be applied to other retinal diseases with a low prevalence.

Optic cup and disc segmentation model based on linear attention and dual attention / 生物医学工程学杂志

Glaucoma is one of blind causing diseases. The cup-to-disc ratio is the main basis for glaucoma screening. Therefore, it is of great significance to precisely segment the optic cup and disc. In this article, an optic cup and disc segmentation model based on the linear attention and dual attention is proposed. Firstly, the region of interest is located and cropped according to the characteristics of the optic disc. Secondly, linear attention residual network-34 (ResNet-34) is introduced as a feature extraction network. Finally, channel and spatial dual attention weights are generated by the linear attention output features, which are used to calibrate feature map in the decoder to obtain the optic cup and disc segmentation image. Experimental results show that the intersection over union of the optic disc and cup in Retinal Image Dataset for Optic Nerve Head Segmentation (DRISHTI-GS) dataset are 0.962 3 and 0.856 4, respectively, and the intersection over union of the optic disc and cup in retinal image database for optic nerve evaluation (RIM-ONE-V3) are 0.956 3 and 0.784 4, respectively. The proposed model is better than the comparison algorithm and has certain medical value in the early screening of glaucoma. In addition, this article uses knowledge distillation technology to generate two smaller models, which is beneficial to apply the models to embedded device.