Automatic 3-D Lamina Curve Extraction From Freehand 3-D Ultrasound Data Using Sequential Localization Recurrent Convolutional Networks

Freehand 3-D ultrasound imaging is emerging as a promising modality for regular spine exams due to its noninvasiveness and affordability. The laminae landmarks play a critical role in depicting the 3-D shape of the spine. However, the extraction of the 3-D lamina curves from transverse ultrasound sequences presents a challenging task, primarily attributed to the presence of diverse contrast variations, imaging artifacts, the complex surface of vertebral bones, and the difficulties associated with probe manipulation. This article proposes sequential localization recurrent convolutional networks (SL-RCNs), a novel deep learning model that takes the contextual relationships into account and embeds the transformation matrix feature as a 3-D knowledge base to enhance accurate ultrasound sequence analysis. The assessment involved the analysis of 3-D ultrasound sequences obtained from ten healthy adult human participants, covering both the lumbar and thoracic regions. The performance of SL-RCN is evaluated through sevenfold cross-validation, using the leave-one-participant-out strategy. The validity of AI model training is assessed on test data from three participants. Normalized discrete Fréchet distance (NDFD) is used as the main metric to evaluate the disparity of the extracted 3-D lamina curves. In contrast to our previous 2-D image analysis method, SL-RCN generates reduced left/right mean distance errors (MDEs) from 1.62/1.63 to 1.41/1.40 mm, and NDFDs from 0.5910/0.6389 to 0.4276/0.4567. The increase in the mean NDFD value from sevenfold cross-validation to the test data experiment is less than 0.05. The experiments demonstrate the SL-RCN’s capability in extracting accurate paired smooth lamina landmark curves.