Ultrasound Image Synthesis Using Generative Ai For Lung Consolidation Detection

Abstract

Lung consolidation remains difficult to diagnose accurately due to the limited availability of large, well-annotated ultrasound datasets, which limits the performance of machine learning models built for clinical support. In order to tackle this challenge, this proposal presents a generative AI-driven framework able to synthesize realistic lung ultrasound images representative of different pathological patterns related to consolidation, aiming at improving the training pipelines by supplying high-fidelity synthetic data that enhances model robustness and generalization. The proposed system combines GAN- and VAE-based generators with MATLAB-based classification pipelines, ensuring that the produced synthetic images will be validated against real clinical samples for fidelity in structure and texture. The novelty of the present work lies in its combined use of several generative architectures for ultrasound realism, its integration into end-to-end ML workflows, and its demonstrated capacity to reduce overfitting while improving the diagnostic accuracy of models in the development of reliable ultrasound-based decision support.