Search

Efficient detection of thin objects, from stationary or moving images, is significant in a variety of research areas. These research areas include but are not limited to electric power line detection systems, sperm tail detection for clinical sperm research, mooring lines detection, road-lane line detection for autonomous vehicles, and cracks detection for the integrity assessment of building structures. However, the detection of thin objects is a challenging computer vision task owing to the slimmer and less compact nature of these objects. Moreover, the complexity present in certain images, such as the background clutter, further adds to this problem of accurately detecting thin objects. In this work, we investigate a series of state-of-the-art deep learning detectors for thin objects’ detection. The detectors examined in this work were: EfficientDet, YOLOv5 and U-Net. The experimental results of this study reveal that generic state-of-the-art deep detectors are not suitable for detecting thin objects due to their reliance on coarse bounding boxes and/or excessive pixel-level computations while the application-specific detectors possess poor generalization capabilities and do not work accurately outside their domains. These empirical findings indicate the necessity of the identification of critical factors affecting thin objects detection and the subsequent design of a generic thin objects’ detector.

This paper investigates the classification of normal and abnormal radiographic images. Eleven convolutional neural network architectures (GoogleNet, Vgg-19, AlexNet, SqueezeNet, ResNet-18, Inception-v3, ResNet-50, Vgg-16, ResNet-101, DenseNet-201 and Inception-ResNet-v2) were used to classify a series of x-ray images from Stanford Musculoskeletal Radiographs (MURA) dataset corresponding to the wrist images of the data base. For each architecture, the results were compared against the known labels (normal / abnormal) and then the following metrics were calculated: accuracy (labels correctly classified) and Cohen's kappa (a measure of agreement) following MURA guidelines. Numerous experiments were conducted by changing classifiers (Adam, Sgdm, RmsProp), the number of epochs, with/without data augmentation. The best results were provided by InceptionResnet-v2 (Mean accuracy = 0.723, Mean Kappa = 0.506). Interestingly, these results lower than those reported in the Leaderboard of MURA. We speculate that to improve the results from basic CNN architectures several options could be tested, for instance: pre-processing, post-processing or domain knowledge, and ensembles.

Cell migration is important in many human processes of development and disease. In Cancer, migration can be related to metastasis or cell defects. A precise analysis of the cell shapes in biological studies could lead to insights about migration. Therefore, this paper describes an algorithm to iteratively segment, track and analyse the shape of macrophages from fluorescent microscopy image sequences. This process allows observation of shape variations as the cells migrate. The algorithm identifies and separates overlapping and non-overlapping cells, then for the non-overlapping cases analyses the shape and extracts a series of measurements, including the number of "corner" or pointy edges through a multiscale angle variation matrix, anglegram. The shape evolution algorithm was tested on fluorescently labelled macrophages observed on embryos of Drosophila melanogaster.

Objective: Ultrawideband radar imaging is regarded as one of the most promising alternatives for breast cancer detection. A range of algorithms reported in literature shows satisfactory tumor detection capabilities. However, most of algorithms suffer significant deterioration or even fail when the early-stage artifact, including incident signals and skin-fat interface reflections, cannot be perfectly removed from received signals. Furthermore, fibro-glandular tissue poses another challenge for tumor detection, due to the small dielectric contrast between glandular and cancerous tissues. Methods: This paper introduces a novel Robust and Artifact Resistant (RAR) algorithm, in which a neighborhood pairwise correlation-based weighting is designed to overcome the adverse effects from both artifact and glandular tissues. In RAR, backscattered signals are time-shifted, summed, and weighted by the maximum combination of the neighboring pairwise correlation coefficients between shifted signals, forming the intensity of each point within an imaging area. Results: The effectiveness was investigated using 3-D anatomically and dielectrically accurate finite-difference-time-domain numerical breast models. The use of neighborhood pairwise correlation provided robustness against artifact and enabled the detection of multiple scatterers. RAR is compared with four well-known algorithms: delay-and-sum, delay-multiply-and-sum, modified-weighted-delay-and-sum, and filtered-delay-and-sum. Conclusion: It has shown that RAR exhibits improved identification capability, robust artifact resistance, and high detectability over its counterparts in most scenarios considered, while maintaining computational efficiency. Simulated tumors in both homogeneous and heterogonous, from mildly to moderately dense breast phantoms, combining an entropy-based artifact removal algorithm, were successfully identified and localized. Significance: These results show the strong potential of RAR for breast cancer screening.

This work describes an automatic algorithm for the segmentation and quantification of focal adhesions from mouse embryonic fibroblasts. The main challenges solved by this algorithm are: the variability of the intensity of the focal adhesions, the detection of an outer ring, which distinguishes the cell periphery responsible for the cell migration, and the quantification of the characteristics of the focal adhesions. The algorithm detects maximal regions through gradients and uses a region-growing algorithm limited by intensity-based edges. The outer ring is calculated based on the average radial intensity from an extended centroid of the cell. Finally, traditional morphological characteristics are obtained to distinguish between two groups of cells. Two of the measurements employed showed statistical difference between two groups of cells.

The aim of this work is to register serial in-vivo confocal microscopy images of zebrafish to enable accurate cell tracking on corresponding fluorescence images. The following problem arises during acquisition; the zebrafish tail may undergoe a series of movement and non-linear deformations, which if not corrected, adds to the motion of leukocytes being tracked. This makes it difficult to accurately assess their motion. We developed a correlation based, local affine image matching method, which is well suited to the textured DIC images of the anatomy of the zebrafish and enables accurate and efficient tracking of image regions over successive frames. Experimental results of the serial registration and tracking demonstrate its accuracy in estimating local affine motions in ze-brafish sequences.