Pusan National University Researchers Create Innovative Deep Learning Model for Enhanced 3D Medical Imaging

In a remarkable advancement for medical diagnostics, researchers from Pusan National University have developed a pioneering deep learning model named MoGLo-Net. This groundbreaking approach revolutionizes handheld 3D medical imaging by eliminating the need for cumbersome and costly external sensors. The traditional method of ultrasound (US) imaging, which employs ultrasonic sound waves to create real-time images of organs and tissues, is significantly enhanced through this innovative model.

Ultrasound is essential for guiding procedures like biopsies and injections, as well as for monitoring blood vessels dynamically. Often, it’s combined with photoacoustic (PA) imaging, a method where laser light pulses generate sound waves in tissues, optimizing the imaging quality. However, even with these advancements, the limitations of conventional 2D imaging have posed challenges. While some apparatuses can achieve comprehensive 3D imaging, they are usually prohibitively expensive and offer a restricted field of view.

The 3D Freehand method provides a solution by stitching together a series of 2D images captured by sweeping a transducer across the body. Nevertheless, this process faces significant hurdles in accurately tracking the motion of the transducer, typically relying on expensive external sensors that can often yield inaccurate readings.

Addressing these challenges, Professor MinWoo Kim and his team from Pusan National University’s School of Biomedical Convergence Engineering and the Center for Artificial Intelligence Research have crafted a solution with MoGLo-Net. This advanced model utilizes tissue speckle data to automatically track the movement of the transducer, thus aiding in the generation of high-fidelity 3D images from 2D ultrasound scans. According to Prof. Kim, “This model empowers doctors with clearer insights into internal bodily functions and enhances their clinical decision-making capabilities.” Their findings were published in the distinguished journal IEEE Transactions on Medical Imaging.

MoGLo-Net operates on a dual-component structure comprising an encoder, inspired by the ResNet framework, and a motion estimator propelled by Long-Short Term Memory (LSTM) neural networks. This encoder is designed with specialized blocks to establish correlations between sequential images derived from tissue speckle patterns, a unique correlation operation technique. Consequently, it captures both in-plane and out-of-plane movements effectively. Notably, the encoder features an innovative self-attention mechanism that draws attention to local characteristics within images, informed by overarching global insights throughout the image.

The model's LSTM-based motion estimator meticulously evaluates transducer movement over time, benefiting from its long-term memory capabilities. Furthermore, the researchers have integrated customized loss functions integral to improving accuracy across various conditions. Testing MoGLo-Net against both proprietary and public datasets revealed its superiority over existing models, delivering more realistic three-dimensional ultrasound imagery.

For the first time in this domain, the team has also successfully merged ultrasound and photoacoustic data to reconstruct 3D images of blood vessels, showcasing a significant leap in ultrasound technology. This major innovation stands to improve access to quality healthcare imaging solutions, making it more accurate, effective, and affordable to diverse patient populations around the globe. With this pioneering work, Pusan National University is not only breaking ground in medical imaging but also setting a new benchmark for future research and applications in the field.

The original paper titled “Enhancing Free-hand 3D Photoacoustic and Ultrasound Reconstruction using Deep Learning” was published on June 13, 2025, in the reputable IEEE Transactions on Medical Imaging journal. Through the continuous pursuit of advancements such as these, the healthcare sector is on the cusp of truly revolutionary diagnostic capabilities.

Topics Health)

【About Using Articles】

You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.

【About Links】

Links are free to use.