Guilherme C. S. Ruppert

A new symmetry-based method for mid-sagittal plane extraction in neuroimages

The estimation of the mid-sagittal plane (MSP) is a known problem with several applications in neuroimage analysis. As advance to the state-of-the-art, we present a considerably better approach for MSP extraction based on bilateral symmetry maximization and a more suitable error metric to compare MSP estimation methods. The proposed method was quantitatively evaluated using three other state-of-the-art approaches as baselines and a heterogeneous dataset with 164 clinical images. It outperformed the others in accuracy and precision, being well succeeded on all images. Besides, it does not present limitations with respect to the imaging protocol and initial position of the head, and it is one of the fastest methods in the literature, taking around 30 seconds on a regular workstation.

How to fix any 3D segmentation interactively via Image Foresting Transform and its use in MRI brain segmentation

In medical imaging, there are many approaches for automatic segmentation. However, none of these methods provide any effective solution to correct segmentation interactively, which becomes a necessity in the case of poorly defined structures. Manual segmentation can not be an alternative given that it might be unfeasible in many cases. On the other hand, how to complete a poor automatic segmentation in an interactive tool is an issue, since the automatic approach and the tool may have been designed with different optimization criteria. We propose solutions to this problem using the framework of the “Image Foresting Transform” (IFT), with evaluation in the context of the segmentation of MR-T1 brain structures. The results indicate that effective semi-automatic correction is possible using just a few markers.

3D Scanning Using RGBD Imaging Devices: A Survey

The capture and digital reconstruction of tridimensional objects and scenarios are issues of great importance in computational vision and computer graphics, for the numerous applications, from navigation and scenario mapping, augmented reality to medical prototyping. In the past years, with the appearance of portable and low-cost devices such as the Kinect Sensor, which are capable of acquiring RGBD video (depth and color data) in real-time, there was a major interest to use these technologies, efficiently, in 3D surface scanning. In this paper, we present a survey of the most relevant methods from recent literature on scanning 3D surfaces using these devices and give the reader a general overview of the current status of the field in order to motivate and enable other works in this topic.

3D blob based brain tumor detection and segmentation in MR images

Automatic detection and segmentation of brain tumors in 3D MR neuroimages can significantly aid early diagnosis, surgical planning, and follow-up assessment. However, due to diverse location and varying size, primary and metastatic tumors present substantial challenges for detection. We present a fully automatic, unsupervised algorithm that can detect single and multiple tumors from 3 to 28,079 mm3 in volume. Using 20 clinical 3D MR scans containing from 1 to 15 tumors per scan, the proposed approach achieves between 87.84% and 95.30% detection rate and an average end-to-end running time of under 3 minutes. In addition, 5 normal clinical 3D MR scans are evaluated quantitatively to demonstrate that the approach has the potential to discriminate between abnormal and normal brains.

Fast, Accurate and Precise Mid-Sagittal Plane Location in 3D MR Images of the Brain

Extraction of the mid-sagittal plane (MSP) is a key step for brain image registration and asymmetry analysis. We present a fast MSP extraction method for 3D MR images, based on automatic segmentation of the brain and on heuristic maximization of the cerebro-spinal fluid within the MSP. The method is robust to severe anatomical asymmetries between the hemispheres, caused by surgical procedures and lesions. The method is also accurate with respect to MSP delineations done by a specialist. The method was evaluated on 64 MR images (36 pathological, 20 healthy, 8 synthetic), and it found a precise and accurate approximation of the MSP in all of them with a mean time of 60.0 seconds per image, mean angular variation within a same image (precision) of 1.26o and mean angular difference from specialist delineations (accuracy) of 1.64o.

Malicious Software Classification Using VGG16 Deep Neural Network’s Bottleneck Features

Malicious software (malware) has been extensively employed for illegal purposes and thousands of new samples are discovered every day. The ability to classify samples with similar characteristics into families makes possible to create mitigation strategies that work for a whole class of programs. In this paper, we present a malware family classification approach using VGG16 deep neural network’s bottleneck features. Malware samples are represented as byteplot grayscale images and the convolutional layers of a VGG16 deep neural network pre-trained on the ImageNet dataset is used for bottleneck features extraction. These features are used to train a SVM classifier for the malware family classification task. The experimental results on a dataset comprising 10,136 samples from 20 different families showed that our approach can effectively be used to classify malware families with an accuracy of 92.97%, outperforming similar approaches proposed in the literature which require feature engineering and considerable domain expertise.

Detecting Computer Generated Images with Deep Convolutional Neural Networks

Computer graphics techniques for image generation are living an era where, day after day, the quality of produced content is impressing even the more skeptical viewer. Although it is a great advance for industries like games and movies, it can become a real problem when the application of such techniques is applied for the production of fake images. In this paper we propose a new approach for computer generated images detection using a deep convolutional neural network model based on ResNet-50 and transfer learning concepts. Unlike the state-of-the-art approaches, the proposed method is able to classify images between computer generated or photo generated directly from the raw image data with no need for any pre-processing or hand-crafted feature extraction whatsoever. Experiments on a public dataset comprising 9700 images show an accuracy higher than 94%, which is comparable to the literature reported results, without the drawback of laborious and manual step of specialized features extraction and selection.

1515 TOUCHLESS GESTURE USER INTERFACE FOR INTERACTIVE IMAGE VISUALIZATION IN UROLOGICAL SURGERY

Purpose In the operating room (OR) a touchless interface is an ideal solution since it does not demand any physical contact and still can provide the necessary control features in a cleansed and sterilized environment.