A Survey of Deep Learning Based Fully Automatic Bone Age Assessment Algorithms

Fine-grained precise-bone age assessment by integrating prior knowledge and recursive feature pyramid network

“…The North American Radiology Society (RSNA) has hosted the famous Pediatric Bone Age Machine Learning Competition Challenge [28], which attracts a large number of researchers to participate in the competition and greatly promotes the development of automatic BAA technology. In recent years, our research group has also jointly carried out several research works with the Xi'an Honghui hospital and the Second Affiliated Hospital of Xi'an Jiaotong University, and published a series of research papers [11][12][13][14][15]. The general framework of automatic BAA nowadays is based on the deep learning method [3,10,[28][29][30].…”

“…In recent years, with the development of deep learning technology, numerous end-to-end BAA algorithms and models based on the deep neural network have emerged. In these end-to-end models, users input an X-ray image into the model and get the bone age as the output directly [9][10][11][12][13][14][15][16][17][18]. Imagining that, once you take an X-ray image, you will get a bone age within several seconds, which will save a lot of time and labor.…”

“…However, clinicians and radiologists think that the scoring method should provide more fine-grained and precise results, and they want to see the development of each bone in hand, rather than just get one single bone age number. If the software can automatically mark the bone development changes of a patient in time sequence, it is much more valuable to analyze the patient's progression and adjust the treatment plan [15]. In order to get the fine-grained and precise bone maturity evaluation result, all of the regions of interest (ROIs) in hand based on the medical standards, such as TW method and RUS-CHN [5,6], need to be detected and analyzed one by one and this precise analysis is more challenging because of the multiple small bone detection task and the lack of precisely annotated ROIs.…”