A carregar...
10 resultados
Resultados da pesquisa
A mostrar 1 - 10 de 10
- Skin lesion classification enhancement using border-line features – The melanoma vs nevus problemPublication . Pereira, Pedro M. M.; Fonseca-Pinto, Rui; Paiva, Rui Pedro; Assuncao, Pedro A. A.; Tavora, Luis M. N.; Thomaz, Lucas A.; Faria, Sergio M. M.Machine learning algorithms are progressively assuming an important role as a computational tool to support clinical diagnosis, namely in the classification of pigmented skin lesions. The current classification methods commonly rely on features derived from shape, colour, or texture, obtained after image segmentation, but these do not always guarantee the best results. To improve the classification accuracy, this work proposes to further exploit the border-line characteristics of the lesion segmentation mask, by combining gradients with local binary patterns (LBP). In the proposed method, these border-line features are used together with the conventional ones to enhance the performance of skin lesion classification algorithms. When the new features are combined with the classical ones, the experimental results show higher accuracy, which impacts positively the overall performance of the classification algorithms. As the medical image datasets usually present large class imbalance, which results in low sensitivity for the classifiers, the border-line features have a positive impact on this classification metric, as evidenced by the experimental results. Both the features’ usefulness and their impact are assessed in regard to the classification results, which in turn are statistically tested for completeness, using three different classifiers and two medical image datasets.
- Integer DCT Approximation With Arbitrary Size and Adjustable PrecisionPublication . Thomaz, Lucas A.; Assunção, Pedro A. A.; Tavora, Luis M. N.; Faria, Sérgio M. M. deThis letter proposes a method to obtain integer reversible discrete cosine transforms for generic transform-based coding schemes. The novelty of the proposed method, which is based on decomposition of the DCT-II matrix into two triangular and one diagonal matrices, is twofold: (i) the new matrices can be of arbitrary size, i.e., any square N\times N dimension, thus suitable for applications where non power-of-2 dimensions are required; (ii) they can be designed with adjustable precision in a trade-off with the number of representation bits. Furthermore, improvements are also proposed over the base scheme to avoid numerical issues when working with large matrices and to obtain more reliable approximations. The performance evaluation demonstrate the effectiveness of the proposed transforms to approximate the coding gain capabilities of the original DCT-II.
- Robust Depth Estimation From Multi-Focus Plenoptic ImagesPublication . Cunha, Francisco; Thomaz, Lucas; Tavora, Luis M. N.; Assunção, Pedro A. A.; Fonseca-Pinto, Rui; Faria, Sérgio M. M.This paper describes a robust depth estimation algorithm for multi-focus plenoptic images. The main feature of the proposed method consists of a hybrid template matching scheme built-upon intensity and local phase information, which adapts to the blurriness of neighbouring lenslet microimages. By reducing the impact of defocusblur on the template matching accuracy, the proposed method efficiently handles the varying triangulation baseline over the depth-of-field, thus discarding the need for scene-related information such as the expected range of disparities. Experimental results demonstrate the robustness of the proposed method over the most used commercially available depth estimation algorithm, achieving a reduction of 73% on the depth estimation error.
- 4D Light Field Disparity Map estimation using Krawtchouk PolynomialsPublication . Lourenco, Rui; Rivero-Castillo, Daniel; Thomaz, Lucas A.; Assuncao, Pedro A. A.; Tavora, Luis M. N.; Faria, Sergio M. M. deThis work presents an improved method to estimate disparity maps obtained from light field cameras using a novel edge detection algorithm based on Krawtchouk polynomials. The proposed method takes advantage of these polynomials to determine gradient information and find the edges based on automatically estimated weak and strong thresholds. The calculated edges in the gray scale epipolar plane image representation of a light field are then used to improve the accuracy of object boundaries in the the disparity map. The proposed method achieves better results when compared to other edge detection algorithms, both in terms of objective and subjective quality, specifically by reducing the mean squared error and the artifacts in the object boundaries. Furthermore, on average, the proposed method outperforms the state-of-the-art depth estimation algorithms, in terms of the objective quality of the final disparity map, namely for the commonly used HCI dataset.
- Disparity compensation of light fields for improved efficiency in 4D transform-based encodersPublication . Santos, Joao M.; Thomaz, Lucas A.; Assuncao, Pedro A. A.; Cruz, Luis A. da Silva; Tavora, Luis M. N.; Faria, Sergio M. M. deEfficient light field en coders take advantage of the inherent 4D data structures to achieve high compression performance. This is accomplished by exploiting the redundancy of co-located pixels in different sub-aperture images (SAIs) through prediction and/or transform schemes to find a m ore compact representation of the signal. However, in image regions with higher disparity between SAIs, such scheme's performance tends to decrease, thus reducing the compression efficiency. This paper introduces a reversible pre-processing algorithm for disparity compensation that operates on the SAI domain of light field data. The proposed method contributes to improve the transform efficiency of the encoder, since the disparity-compensated data presents higher correlation between co-located image blocks. The experimental results show significant improvements in the compression performance of 4D light fields, achieving Bjontegaard delta rate gains of about 44% on average for MuLE codec using the 4D discrete cosine transform, when encoding High Density Camera Arrays (HDCA) light field images.
- Dermoscopic skin lesion image segmentation based on Local Binary Pattern Clustering: Comparative studyPublication . Pereira, Pedro M. M.; Fonseca-Pinto, Rui; Paiva, Rui Pedro; Assuncao, Pedro A. A.; Tavora, Luis M. N.; Thomaz, Lucas A.; Faria, Sergio M. M.Accurate skin lesion segmentation is important for identification and classification through computational methods. However, when performed by dermatologists, the results of clinical segmentation are affected by a certain margin of inaccuracy (which exists since dermatologist do not delineate lesions for segmentation but for extraction) and also significant inter- and intra-individual variability, such segmentation is not sufficiently accurate for segmentation studies. This work addresses these limitations to enable detailed analysis of lesions’ geometry along with extraction of non-linear characteristics of region-of-interest border lines. A comprehensive review of 39 segmentation methods is carried out and a contribution to improve dermoscopic image segmentation is presented to determine the regions-of-interest of skin lesions, through accurate border lines with fine geometric details. This approach resorts to Local Binary Patterns and k-means clustering for precise identification of lesions boundaries, particularly the melanocytic. A comparative evaluation study is carried out using three different datasets and reviewed algorithms are grouped according to their approach. Results show that algorithms from the same group tend to perform similarly. Nevertheless, their performance does not depend uniquely on the algorithm itself but also on the underlying dataset characteristics. Throughout several evaluations, the proposed Local Binary Patterns method presents, consistently, better average performance than the current state-of-the-art techniques across the three different datasets without the need of training or supervised learning steps. Overall, apart from presenting a new segmentation method capable of outperforming the current state-of-the-art, this paper provides insightful information about the behaviour and performance of different image segmentation algorithms.
- Explainable prototype-based image classification using adaptive feature extractors in medical imagesPublication . Vasconcellos, Nicolas; Tavora, Luis M. N.; Miragaia, Rolando; Grilo, Carlos; Thomaz, LucasPrototype-based classifiers are a category of Explainable Artificial Intelligence methods that use representative samples from the data, called prototypes, to classify new inputs based on a similarity criterion. However, these methods often rely on pre-trained Convolutional Neural Networks as feature extractors, which may not be adapted for the specific type of data being used, thus not suited for identifying the most representative prototypes. In this paper, we propose a method named Explainable Prototype-based Image Classification, a cluster-oriented training strategy that enhances the performance and explainability of prototype-based classifiers. Our method uses a novel loss function, called Cluster Density Error, to fine-tune the feature extractor and preserve the most representative feature vectors in the latent space. We also use Principal Component Analysis-based approach to reduce the dimensionality and complexity of the feature vectors. We conduct experiments on four medical image datasets and compare the results with those from different prototype-based classifiers and state-of-the-art non-explainable learning methods. The proposed method demonstrated superior explainable capabilities and comparable classification performance to the compared methods. Specifically, the proposed method achieved up to 95.01% accuracy and 0.992 AUC using only 43 prototypes. This translated to an improvement in accuracy and AUC score of 21.54% and 9.06%, respectively, and a substantial reduction in the number of prototypes by 98,38%
- Light Field Disparity Map Enhancement with Morphological FilteringPublication . Lourenco, Rui; Thomaz, Lucas A.; Silva, Eduardo A. B. da; Assuncao, Pedro A. A.; Tavora, Luis M. N.; Faria, Sergio M. M. deLight field disparity estimation algorithms are comprised of two steps: an initial estimation step and a global optimization step. The initial estimation is often noisy and may contain high amplitude artefacts. Global optimization techniques might inadequately propagate these artefacts, providing suboptimal results. In this paper, an iterative morphological filter is proposed as an intermediate step or replacement to global optimization techniques. This algorithm iteratively filters the disparity map with an average of Open followed by Close and Close followed by Open morphological operations, enabling the removal of artefacts and noise, without adversely affecting the structure of the disparity map. The iterative open-close close-open filter attenuates the effect of artefacts and noise from an initial disparity estimation, achieving improvements of up to 90%, and more than 30%, on average, in terms of mean square error, when applied to the a structure-tensor-based initial estimation. In addition, the proposed method proves to be competitive with another state of the art algorithm, in terms of mean square error, and superior in terms of percentage of bad pixels.
- Cross-modality Lossless Compression of PET-CT ImagesPublication . Parracho, João O.; Thomaz, Lucas A.; Távora, Luís M. N.; Assunção, Pedro A. A.; Faria, Sérgio M. M.The huge amount of data resulting from the acquisition of medical images with multiple modalities can be overwhelming for storage and sharing through communication systems. Thus, efficient compression algorithms must be introduced to reduce the burden of storage and communication resources required by such amount of data. However, since in the medical context all details are important, the adoption of lossless image compression algorithms is paramount. This paper proposes a novel lossless compression scheme tailored to jointly compress the modality of computerized tomography (CT) and that of positron emission tomography (PET). Different approaches are adopted, namely image-to-image translation techniques and redundancies between both images are also explored. To perform the image-to-image translation approach, we resort to lossless compression of the original CT data and apply a cross-modality image translation generative adversarial network to obtain an estimation of the corresponding PET. Then, the residue that results from the differences between the original PET and its estimation is also compressed. Thus, instead of compressing two independent image modalities, i.e., both images of the original PET-CT pair, in the proposed approach only the CT is independently encoded along with the PET residue. The performed experiments using a publicly available PET-CT pair dataset show that the proposed scheme attains up to 8.9 % compression gains for the PET data, in comparison with the naive approach, and up to 3.5 % gains for the PET-CT pair.
- Skin Lesion Classification using Bag-of-3D-FeaturesPublication . Pereira, Pedro M. M.; Thomaz, Lucas A.; Tavora, Luis M. N.; Assuncao, Pedro A. A.; Fonseca-Pinto, Rui; Paiva, Rui Pedro; Faria, Sergio M. M.Computer-aided diagnostic has become a thriving research area in recent years, namely on the identification of skin lesions such as melanoma. This work presents a novel approach to this field by exploiting the 3D characteristics of the skin lesion surface, advancing beyond common features such as, shape, colour, and texture, extracted from dermoscopic RGB images. To this end, a relevant set of features was investigated to obtain 3D skin lesion characteristics from images with depth information. These features were used to train a Bag-of-Features model to distinguish between malignant and benign lesions, also discriminating melanoma from all other lesion types. Despite the large class imbalance, often present in medical image datasets, the feature set achieved a top accuracy of 73.08%, comprising 75.00% sensitivity and 66.67% specificity when classifying between malignant and benign lesions, and 88.46% accuracy (100.00% sensitivity and 86.96% specificity) when discriminating melanoma from all other lesion images, using only depth information. The achieved experimental results indicate the existence of relevant discriminative characteristics in the 3D surface of skin lesions which allow the improvement of existing classification methods based on 2D image characteristics only.