In optoacoustics, image quality depends on the number and distribution of sensors used by the device; the more sensors and the more broadly they are arranged, the better the quality. To improve image quality in low-cost optoacoustic devices with only a small number of ultrasonic sensors, researchers at ETH Zurich and the University of Zurich turned to machine learning. They developed a framework for the efficient recovery of image quality from sparse optoacoustic data using a deep convolutional neural network and demonstrated their approach with whole body mouse imaging in vivo. To generate accurate, high-resolution reference images for training, the team began by developing a high-end optoacoustic scanner with 512 sensors. An artificial neural network analyzed and learned the features of the superior-quality images that were generated by this device. The researchers then removed most of the sensors from their device, causing the quality of its imaging to degrade. Due to the lack of data, streak-type artifacts appeared in the images. However, the previously trained neural network was able to correct for most of these distortions, bringing the image quality close to the measurements obtained when the device had all 512 sensors. Further, the machine learning algorithm developed by the team was able to improve the quality of images that were recorded from a narrowly circumscribed sector. “This is particularly important for clinical applications, as the laser pulses cannot penetrate the entire human body, hence the imaged region is normally only accessible from one direction,” professor Daniel Razansky said. No comparable gains were achieved when the training was performed with synthetic or phantom data, indicating the importance of training with high-quality in vivo images acquired by full-view scanners. The team said its approach could be applied to other imaging techniques because the approach is based on reconstructed images, not raw recorded data. “You can basically use the same methodology to produce high-quality images from any sort of sparse data,” Razansky said. Physicians are often confronted with the challenge of interpreting poor quality images. “We show that such images can be improved with AI methods, making it easier to attain more accurate diagnosis,” he said. The new method could benefit numerous optoacoustic imaging applications by mitigating common image artifacts, enhancing anatomical contrast and image quantification capacities, and accelerating data acquisition and image reconstruction. It could also facilitate the development of practical, affordable optoacoustic imaging systems. For their current research, the scientists used an optoacoustic tomography device customized for small animals and trained the machine learning algorithms with images from mice. Their next step will be to apply their method to optoacoustic images from human patients. Optoacoustic imaging is especially useful for visualizing blood vessels. Courtesy of ETH Zurich/Daniel Razansky. About Optoacoustics Optoacoustics is similar to ultrasound imaging. In ultrasound imaging, a probe sends ultrasonic waves into the body, which are reflected by the tissue. Sensors in the probe detect the returning sound waves and a picture of the inside of the body is subsequently generated. In optoacoustic imaging, very short laser pulses are sent into the tissue, where they are absorbed and converted into ultrasonic waves. Similarly to ultrasound imaging, the waves are detected and converted into images. Many imaging techniques, such as ultrasound, x-ray, or MRI, are mainly suitable for visualizing anatomical alterations in the body. To receive additional functional information — for instance, concerning blood flow or metabolic changes — the patient has to be administered contrast agents or radioactive tracers before the imaging. In contrast, the optoacoustic method can visualize functional and molecular information without introducing contrast agents. One example is local changes in tissue oxygenation, an important landmark of cancer that can be used for early diagnosis. However, because the lightwaves used in optoacoustic imaging do not fully penetrate the human body, the method is only suitable for investigating tissues to a depth of a few centimeters beneath the skin. The research was published in Nature Machine Intelligence (www.doi.org/10.1038/s42256-019-0095-3).