Purpose: The deep learning-aided automatic bone age assessment systems have obtained the best results but still depending on the advanced methods to attain highly accurate region of interest (ROI) segmentation and error-free bone age assessment based on different regions of bone. By avoiding these drawbacks, this paper presents a comparative analysis on a deep learning-aided bone age assessment using 5 regions over 13 regions under Tanner-Whitehouse 3 (TW3) method. Methods: The segmentation of both 5 regions and 13 regions is employed by both adaptive Otsu thresholding and improved U-Net segmentation. Further, the adoption of enhanced “deep convolutional neural network (CNN) is used for the bone age assessment (BAA)” using the 13 region and 5 regions. The development of Opposition Searched Harris Hawks Optimization (OS-HHO) is preferable for both segmentation and prediction. Results: From the analysis, the root mean square error (RMSE) of 5th region is 8.45% better than that of the 13th region. Similarly, the mean absolute squared error (MASE) of 5th region was 2.30% higher than the 13th region. Conclusion: The assessment of bone age by 5 regions was better when compared with the 13 regions for the suggested bone age assessment (BAA) model, respectively. © 2022, Sociedade Brasileira de Engenharia Biomedica.