TGC-Net: A Deep Learning Model for Automatic Time Gain Compensation for Ultrasound Imaging

The shift towards point-of-care ultrasound (POCUS) has driven innovations in high-quality imaging for low resource settings, driven by artificial intelligence (AI) to improve diagnostics. While research primarily focuses on enhancing imaging through deep neural networks (DNN) for beamforming, reliable ultrasound (US) images also require time gain compensation (TGC) to account for the attenuation of sound waves received from deep regions. The TGC is performed manually by the operator by adjusting the anatomyspecific gain settings; hence, the optimal image quality depends on the skill and expertise of the operator. To address this, we introduce a novel and robust DNN model called TGC-net, which automatically optimizes lateral and axial gains with depth to enhance US images. We train and test the proposed TGC-net model using both simulated and experimental datasets and subsequently validate the performance of the TGC-net through various qualitative and quantitative metrics such as full width half maximum, contrast to noise ratio, generalized contrast to noise ratio, and signal-to-noise ratio. A comprehensive evaluation of the results suggests that the TGC-net model produces superior image quality compared to the ground truth.