Copy-Paste Image Augmentation with Poisson Image Editing for Ultrasound Instance Segmentation Learning

Deep learning has shown great success in high-level image analysis problems; yet its efficacy relies on the quality and diversity of the training data. In this work, we introduce a copypaste image augmentation for ultrasound images. The Poisson image editing technique is used to generate realistic and seamless boundary transitions around the pasted image. Results showed that the proposed image augmentation technique improves training performance in terms of higher objective metrics and more stable training results.

Phase Aberration Correction with Adaptive Coherence-Weighted Point Spread Function Restoration Filtering Technique

Phase aberration is an inherent side effect of ultrasound imaging due to the speed of sound inhomogeneity nature of human tissues, resulting in focusing error and reduced image contrast. This work introduces a phase aberration correction technique by leveraging a point spread function (PSF) restoration filter. A convolutional neural network (CNN) is used to estimate phase-aberrated PSFs and design the restoration filter. In addition, we incorporate coherence index weighting, derived from the restoration filtering, to further suppress sidelobe energy. Evaluation using Field II-simulated phantoms showed clearer cyst borders and reduced sidelobe energy leakage after PSF restoration and filter-derived coherence weighting, leading to improvement in image contrast and quality.

Representative Image Feature Extraction via Contrastive Learning Pretraining for Chest X-ray Report Generation

Medical report generation is a challenging task since it is time-consuming and requires expertise from experienced radiologists. The goal of medical report generation is to accurately capture and describe the image findings. Previous works pretrain their visual encoding neural networks with large datasets in different domains, which cannot learn general visual representation in the specific medical domain. In this work, we propose a medical report generation framework that uses a contrastive learning approach to pretrain the visual encoder and requires no additional meta information. In addition, we adopt lung segmentation as an augmentation method in the contrastive learning framework. This segmentation guides the network to focus on encoding the visual feature within the lung region. Experimental results show that the proposed framework improves the performance and the quality of the generated medical reports both quantitatively and qualitatively.