Breast lesion segmentation in ultrasound images with limited annotated data
This addresses the problem of high annotation costs for medical professionals in ultrasound imaging, though it is incremental as it adapts existing methods to data-scarce scenarios.
The study tackled breast lesion segmentation in ultrasound images with limited annotated data by using simulated and natural images for pre-training, then fine-tuning with as few as 19 in vivo images, resulting in a 21% improvement in dice score compared to training from scratch.
Ultrasound (US) is one of the most commonly used imaging modalities in both diagnosis and surgical interventions due to its low-cost, safety, and non-invasive characteristic. US image segmentation is currently a unique challenge because of the presence of speckle noise. As manual segmentation requires considerable efforts and time, the development of automatic segmentation algorithms has attracted researchers attention. Although recent methodologies based on convolutional neural networks have shown promising performances, their success relies on the availability of a large number of training data, which is prohibitively difficult for many applications. Therefore, in this study we propose the use of simulated US images and natural images as auxiliary datasets in order to pre-train our segmentation network, and then to fine-tune with limited in vivo data. We show that with as little as 19 in vivo images, fine-tuning the pre-trained network improves the dice score by 21% compared to training from scratch. We also demonstrate that if the same number of natural and simulation US images is available, pre-training on simulation data is preferable.