Bone Age Assessment (BAA) from X-ray images plays a crucial role in diagnosing growth disorders in children, as bone age may not always align with chronological age. In addition to being time-consuming, traditional manual assessment techniques are also prone to expert variability. Deep learning offers an automated and accurate alternative, particularly for analyzing left wrist X-ray images. However, challenges such as variations in image quality, limited pediatric datasets, and the need for generalizability across diverse populations persist. To address these issues, this manuscript proposes an Enhanced Bone Age Assessment approach utilizing Attention Spiking Neural Networks and Exponential Particle Swarm Optimization (BAA-ASNN-EPSOA). Initially, X-ray images are sourced from RSNA and Digital Hand Atlas datasets. These images undergo pre-processing using the Feedback Correction-Adaptive Kalman Filter (FCKF), which effectively reduces noise and enhances image quality. The refined images are then processed using Multi-View Intrinsic Subspace Clustering (MISC) to segment the region of interest (ROI) accurately. Following segmentation, the images are analyzed using the Attention Spiking Neural Network (ASNN) to estimate bone age. However, ASNN alone does not incorporate adaptive optimization strategies for selecting optimal parameters. To enhance its performance, the Exponential Particle Swarm Optimization Algorithm (EPSOA) is introduced to optimize the weight parameters of ASNN, ensuring precise bone age estimation. The proposed BAA-ASNN-EPSOA is implemented in Python, its effectiveness is evaluated under some metrics, like Accuracy, Recall, Precision, F1-Score, Mean Absolute Error (MAE), Receiver Operating Characteristic (ROC), Root Mean Square Error (RMSE), Computation Time. Compared to existing methods like BAA-XRI-MMANet, RBA-HBM-CNN, and PBAA-HB-DADPN, the proposed approach achieves significant improvements 23.54%, 21.43%, and 21.76% higher accuracy; 21.28%, 23.52%, and 23.58% higher precision; and 24.62%, 23.51%, and 23.42% higher recall, respectively.
Sasidhar et al. (Mon,) studied this question.