Bone age evaluation method based on convolutional neural network and multiple attention mechanism

Abstract

The invention discloses a bone age evaluation method based on a convolutional neural network and a multiple attention mechanism. The method comprises the following steps: in a training stage, taking ametacarpal bone image as the input of a backbone network, obtaining a feature map F through a feature extractor, and obtaining a bone age regression value; taking the input of a multi-attention module as a feature map F, obtaining M sub-attention maps through compression operation and attention map splitting operation, conducting point multiplication of each sub-attention map with the feature mapF, and then obtaining a corresponding bone age regression value; combining a backbone network with the bone age regression value obtained by the multiple attention module, and training a neural network by adopting a multi-task learning strategy; and in a test stage: inputting a to-be-tested metacarpal bone image into the trained neural network, and obtaining a bone age evaluation value through the backbone network. The model can be trained end to end; meanwhile, an attention distribution diagram can be automatically generated, and better generalization is achieved; and in addition, based on the 2D convolutional neural network, the speed is high, the precision is high, and an average evaluation error is within 4.1 months.

Description

technical field [0001] The invention relates to the technical field of intelligent medical image analysis, in particular to a bone age assessment method based on a convolutional neural network and a multiple attention mechanism. Background technique [0002] The traditional bone age assessment usually takes X-rays of the subject's left palm and wrist, and then conducts bone age assessment with the help of common standards. This process relies heavily on the practitioner's experience and is also time-consuming. In addition, there are great differences in skeletal development under different races, climates and other conditions, so the corresponding standards also vary widely, which increases the complexity of bone age assessment. [0003] In order to speed up the evaluation, improve the accuracy of the evaluation and reduce the work intensity, a computer-aided system (CAD) based on artificial intelligence came into being, and achieved a higher accuracy than human experts in ...

AI Technical Summary

Problems solved by technology

However, most of these artificial intelligence technologies introduce fine-grained region-of-interest annotations, and focus on specific bone parts as regions of interest (such as wrist bones, proximal phalanges, etc.) through detection and segmentation methods, which brings expensive annotations. Cost and model complexity limit the practical application value of the method

Images