Evaluation of the Patient Effective Dose in Whole Spine Scanography Based on the Automatic Image Pasting Method for Digital Radiography

Deep learning-based fully automatic Risser stage assessment model using abdominal radiographs

BACKGROUND

Artificial intelligence has been increasingly used in medical imaging and has demonstrated expert level performance in image classification tasks.

OBJECTIVE

To develop a fully automatic approach for determining the Risser stage using deep learning on abdominal radiographs.

MATERIALS AND METHODS

In this multicenter study, 1,681 supine abdominal radiographs (age range, 9-18 years, 50% female) obtained between January 2019 and April 2022 were collected retrospectively from three medical institutions and graded manually using the United States Risser staging system. A total of 1,577 images from Hospitals 1 and 2 were used for development, and 104 images from Hospital 3 for external validation. From each radiograph, right and left iliac crest patch images were extracted using the pelvic bone segmentation model DeepLabv3 + with the EfficientNet-B0 encoder trained with 90 digitally reconstructed radiographs from pelvic computed tomography scans with a pelvic bone mask. Using these patch images, ConvNeXt-B was trained to grade according to the Risser classification. The model's performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUROC), and mean absolute error.

RESULTS

The fully automatic Risser stage assessment model showed an accuracy of 0.87 and 0.75, mean absolute error of 0.13 and 0.26, and AUROC of 0.99 and 0.95 on internal and external test sets, respectively.

CONCLUSION

We developed a deep learning-based, fully automatic segmentation and classification model for Risser stage assessment using abdominal radiographs.

Usefulness of an Additional Filter Created Using 3D Printing for Whole-Body X-ray Imaging with a Long-Length Detector

We recently developed a long-length detector that combines three detectors and successfully acquires whole-body X-ray images. Although the developed detector system can efficiently acquire whole-body images in a short time, it may show problems with diagnostic performance in some areas owing to the use of high-energy X-rays during whole-spine and long-length examinations. In particular, during examinations of relatively thin bones, such as ankles, with a long-length detector, the image quality deteriorates because of an increase in X-ray transmission. An additional filter is primarily used to address this limitation, but this approach imposes a higher load on the X-ray tube to compensate for reductions in the radiation dose and the problem of high manufacturing costs. Thus, in this study, a newly designed additional filter was fabricated using 3D printing technology to improve the applicability of the long-length detector. Whole-spine anterior–posterior (AP), lateral, and long-leg AP X-ray examinations were performed using 3D-printed additional filters composed of 14 mm thick aluminum (Al) or 14 mm thick Al + 1 mm thick copper (Cu) composite material. The signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and radiation dose for the acquired X-ray images were evaluated to demonstrate the usefulness of the filters. Under all X-ray inspection conditions, the most effective data were obtained when the composite additional filter based on a 14 mm thick Al + 1 mm thick Cu material was used. We confirmed that an SNR improvement of up to 46%, CNR improvement of 37%, and radiation dose reduction of 90% could be achieved in the X-ray images obtained using the composite additional filter in comparison to the images obtained with no filter. The results proved that the additional filter made with a 3D printer was effective in improving image quality and reducing the radiation dose for X-ray images obtained using a long-length detector.