.jpg?format=webp&width=780)
The study aimed to compare the effectiveness of supervised learning (SL) and semi-supervised learning (SSL) in classifying mandibular third molars (Mn3s) on panoramic images. A total of 1625 Mn3s images were labeled for the depth of impaction, spatial relation with the second molar, and relationship with inferior alveolar nerve canal. The WideResNet (WRN) model was used for SL, and the LaplaceNet (LN) model was utilized for SSL. The results showed that the SSL model with only 40 labeled images achieved prediction accuracy similar to the SL model using 300-360 labeled images, demonstrating the effectiveness of SSL in classification tasks.
Like
Save
Share