Estimating body ratios from X-rays is an interesting research topic important for clinical applications such as assessing children’s development or deviations from expected growth patterns. Research on these issues typically involves time-consuming and resource-intensive methods, such as manual body length measurements or high-resolution image annotations. We have proposed a preprocessing framework and adapted pre-trained convolutional neural network models to estimate the lengths of the body parts of children and young adults from low-resolution, dual-energy, whole-body, X-ray absorptiometry images. Furthermore, we proposed a solution for the automatic estimation of body ratios and the determination of samples significantly deviating from the average. The dataset of X-ray absorptiometry images was expanded with annotations of the length of individual body parts performed under the supervision of a specialist. The experimental results show that the proposed preprocessing techniques and the adapted convolutional neural network model achieved a mean relative error of up to 4.24 between the estimated and annotated lengths of body parts percent and the Lin correlation coefficient of convergence of 0.92. It was shown that, by employing low-resolution images and deep learning methods, one can accurately estimate the lengths of specific body parts, facilitating the precise detection of samples that deviate from the average.