Search

The automatic generation of accurate radiology reports is of great clinical importance and has drawn growing research interest. However, it is still a challenging task due to the imbalance between normal and abnormal descriptions and the multi-sentence and multi-topic nature of radiology reports. These features result in significant challenges to generating accurate descriptions for medical images, especially the important abnormal findings. Previous methods to tackle these problems rely heavily on extra manual annotations, which are expensive to acquire. We propose a multi-grained report generation framework incorporating sentence-level image-sentence contrastive learning, which does not require any extra labeling but effectively learns knowledge from the image-report pairs. We first introduce contrastive learning as an auxiliary task for image feature learning. Different from previous contrastive methods, we exploit the multi-topic nature of imaging reports and perform fine-grained contrastive learning by extracting sentence topics and contents and contrasting between sentence contents and refined image contents guided by sentence topics. This forces the model to learn distinct abnormal image features for each specific topic. During generation, we use two decoders to first generate coarse sentence topics and then the fine-grained text of each sentence. We directly supervise the intermediate topics using sentence topics learned by our contrastive objective. This strengthens the generation constraint and enables independent fine-tuning of the decoders using reinforcement learning, which further boosts model performance. Experiments on two large-scale datasets MIMIC-CXR and IU-Xray demonstrate that our approach outperforms existing state-of-the-art methods, evaluated by both language generation metrics and clinical accuracy.

Storage of video information is experiencing rapid growth as digital broadcasting services have developed. This has rapidly increased video data storage requirements in DVD and hard disc recorders in recent years. We design a video summarization tool for personal video recorders. This tool provides a video summarization function to grasp the original long video content quickly. This tool is with a two-level redundancy detection procedure. After the video parsing step, it removes redundant video content with the hierarchical agglomerative cluster method in the shot level. Impact factors of scenes and key frames are defined, and parts of shots are selected to generate the initial video summary. Finally a repetitive frame segment detection step is used to remove redundant information. We have developed a prototype model and tested this tool on TV series, movies and cartoons data. The experiment results are given to illustrate the proposed tool. (1) Index Terms--Video Summarization, Redundant Detection, Personal Video Recorder.

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.compind.2006.06.002 Byline: Xu-Zheng Liu (a)(b), Jun-Hai Yong (a), Guo-Qin Zheng (a), Jia-Guang Sun (a)(b) Keywords: Polyline curve; Offset curve; Clipping algorithm Abstract: Polyline curves which are composed of line segments and arcs are widely used in engineering applications. In this paper, a novel offset algorithm for polyline curves is proposed. The offset algorithm comprises three steps. Firstly, the offsets of all the segments of polyline curves are calculated. Then all the offsets are trimmed or joined to build polyline curves that are called untrimmed offset curves. Finally, a clipping algorithm is applied to the untrimmed offset curves to yield the final results. The offset algorithm can deal with polyline curves that are self-intersection, overlapping or containing small arcs. The new algorithm has been implemented in a commercial system TiOpenCAD 8.0 and its reliability is verified by a great number of examples. Author Affiliation: (a) School of Software, Tsinghua University, Beijing 100084, PR China (b) Department of Computer Science and Technology, Tsinghua University, Beijing 100084, PR China Article History: Received 22 July 2004; Accepted 6 June 2006

Deep learning has yielded promising results for medical image diagnosis but relies heavily on manual image annotations, which are expensive to acquire. We present Cross-DL, a cross-modality learning framework for intracranial abnormality detection and localization in head computed tomography (CT) scans by learning from free-text imaging reports. Cross-DL has a discretizer that automatically extracts discrete labels of abnormality types and locations from reports, which are utilized to train an image analyzer by a dynamic multi-instance learning approach. Benefiting from the low annotation cost and a consequent large-scale training set of 28,472 CT scans, Cross-DL achieves accurate performance, with an average area under the receiver operating characteristic curve (AUROC) of 0.956 (95% confidence interval: 0.952–0.959) in detecting 4 abnormality types in 17 regions while accurately localizing abnormalities at the voxel level. An intracranial hemorrhage classification experiment on the external dataset CQ500 achieves an AUROC of 0.928 (0.905–0.951). The model can also help review prioritization.

Point projection on an implicit surface is essential for the geometric modeling and graphics applications of it. This paper presents a method for computing the principle curvatures and principle directions of an implicit surface. Using the principle curvatures and principle directions, we construct a torus patch to approximate the implicit surface locally. The torus patch is second order osculating to the implicit surface. By taking advantage of the approximation torus patch, this paper develops a second order geometric iterative algorithm for point projection on the implicit surface. Experiments illustrate the efficiency and less dependency on initial values of our algorithm.

A method for computing the visible regions of free-form surfaces is proposed in this paper. Our work is focused on accurately calculating the visible regions of the sequenced rational Bézier surfaces forming a solid model and having coincident edges but no inner-intersection among them. The proposed method calculates the silhouettes of the surfaces without tessellating them into triangle meshes commonly used in previous methods so that arbitrary precision can be obtained. The computed silhouettes of visible surfaces are projected onto a plane orthogonal to the parallel light. Then their spatial relationship is applied to calculate the boundaries of mutual-occlusion regions. As the connectivity of the surfaces on the solid model is taken into account, a surface clustering technique is also employed and the mutual-occlusion calculation is accelerated. Experimental results showed that our method is efficient and robust, and can also handle complex shapes with arbitrary precision.

A new method for smoothly connecting different patches on triangle meshes with arbitrary connectivity, called mesh blending, is presented. A major feature of mesh blending is to move vertices of the blending region to a virtual blending surface by choosing an appropriate parameterization of those vertices. Once blending is completed, the parameterization optimization is performed to perfect the final meshes. Combining mesh blending with multiresolution techniques, an effective blending technique for meshes is obtained. Our method has several advantages: (1) the user can intuitively control the blending result using different blending radii, (2) the shape of cross-section curves can be adjusted to flexibly design complex models, and (3) the resulting mesh has the same connectivity as the original mesh. In this paper, some examples about smoothing, sharpening, and mesh editing show the efficiency of the method.

Intracranial aneurysm (IA) is an enormous threat to human health, which often results in nontraumatic subarachnoid hemorrhage or dismal prognosis. Diagnosing IAs on commonly used computed tomographic angiography (CTA) examinations remains laborious and time consuming, leading to error-prone results in clinical practice, especially for small targets. In this study, we propose a fully automatic deep-learning model for IA segmentation that can be applied to CTA images. Our model, called Global Localization-based IA Network (GLIA-Net), can incorporate the global localization prior and generates the fine-grain three-dimensional segmentation. GLIA-Net is trained and evaluated on a big internal dataset (1,338 scans from six institutions) and two external datasets. Evaluations show that our model exhibits good tolerance to different settings and achieves superior performance to other models. A clinical experiment further demonstrates the clinical utility of our technique, which helps radiologists in the diagnosis of IAs.