Search

Background and objectives: The scarcity of data for training deep learning models in pediatrics has prompted questions about the feasibility of employing CNNs trained with adult images for pediatric populations. In this work, a pneumonia classification CNN was used as an exploratory example to showcase the adaptability and efficacy of such models in pediatric healthcare settings despite the inherent data constraints. Methods: To develop a curated training dataset with reduced biases, 46,947 chest X-ray images from various adult datasets were meticulously selected. Two preprocessing approaches were tried to assess the impact of thoracic segmentation on model attention outside the thoracic area. Evaluation of our approach was carried out on a dataset containing 5,856 chest X-rays of children from 1 to 5 years old. Results: An analysis of attention maps indicated that networks trained with thorax segmentation placed less attention on regions outside the thorax, thus eliminating potential bias. The ensuing network exhibited impressive performance when evaluated on an adult dataset, achieving a pneumonia discrimination AUC of 0.95. When tested on a pediatric dataset, the pneumonia discrimination AUC reached 0.82. Conclusions: The results of this study show that adult-trained CNNs can be effectively applied to pediatric populations. This could potentially shift focus towards validating adult models over pediatric population instead of training new CNNs with limited pediatric data. To ensure the generalizability of deep learning models, it is important to implement techniques aimed at minimizing biases, such as image segmentation or low-quality image exclusion. [ABSTRACT FROM AUTHOR]

Image-to-image translation from a source to a target domain by means of generative adversarial neural network (GAN) has gained a lot of attention in the medical imaging field due to their capability to learn the mapping characteristics between different modalities. CycleGAN has been proposed for image-to-image translation with unpaired images by means of a cycle-consistency loss function, which is optimized to reduce the difference between the image reconstructed from the synthetically-generated domain and the original input. However, CycleGAN inherently implies that the mapping between both domains is invertible, i.e., given a mapping G (forward cycle) from domain A to B, there is a mapping F (backward cycle) that is the inverse of G. This is assumption is not always true. For example, when we want to learn functional activity from structural modalities. Although it is well-recognized the relation between structure and function in different physiological processes, the problem is not invertible as the original modality cannot be recovered from a given functional response. In this paper, we propose a functional-consistent CycleGAN that leverages the usage of a proxy structural image in a third domain, shared between source and target, to help the network learn fundamental characteristics while being cycle consistent. To demonstrate the strength of the proposed strategy, we present the application of our method to estimate iodine perfusion maps from contrast CT scans, and we compare the performance of this technique to a traditional CycleGAN framework.