Data Collection and Dataset Construction for Skull X-ray Images

Data Collection Study Utilizes Deep Learning Models for Skull X-ray Analysis

A recent study conducted at Hallym University Sacred Hospital focused on analyzing plain skull X-ray images of infants under 12 months of age who underwent head trauma evaluation. The study, approved by the Institutional Review Board, aimed to develop deep learning models for accurate age prediction based on these images.

The dataset included images taken using a digital radiographic device and retrieved in DICOM format. Personal information was removed to ensure patient confidentiality, and only properly focused images were included. The dataset was divided into training, validation, and test subsets for model development.

To enhance accuracy, the images were pre-processed to hide teeth and sinus areas, and a region of exclusion was identified by a neurosurgical expert. The dataset was then used to train two different CNN models, DenseNet-121 and EfficientNet-V2-M, known for their efficiency in medical image classification.

The models were fine-tuned and trained on a PyTorch platform using GPU hardware. The performance of the models was evaluated based on classification accuracy and one-month relaxation accuracy. The study also compared the proposed method with the RSNA Bone Challenge Winner Model to validate its superiority.

Overall, the study demonstrated the potential of deep learning models in accurately predicting the age of infants based on skull X-ray images. The ethical standards were followed, and informed consent was waived due to the retrospective nature of the study.

With the increasing use of deep learning in medical imaging, studies like this pave the way for more accurate and efficient diagnostic tools in healthcare.