Eduardo Romero

A semi-automatic method for quantification and classification of erythrocytes infected with malaria parasites in microscopic images

Visual quantification of parasitemia in thin blood films is a very tedious, subjective and time-consuming task. This study presents an original method for quantification and classification of erythrocytes in stained thin blood films infected with Plasmodium falciparum. The proposed approach is composed of three main phases: a preprocessing step, which corrects luminance differences. A segmentation step that uses the normalized RGB color space for classifying pixels either as erythrocyte or background followed by an Inclusion-Tree representation that structures the pixel information into objects, from which erythrocytes are found. Finally, a two step classification process identifies infected erythrocytes and differentiates the infection stage, using a trained bank of classifiers. Additionally, user intervention is allowed when the approach cannot make a proper decision. Four hundred fifty malaria images were used for training and evaluating the method. Automatic identification of infected erythrocytes showed a specificity of 99.7% and a sensitivity of 94%. The infection stage was determined with an average sensitivity of 78.8% and average specificity of 91.2%.

Single-image super-resolution of brain MR images using overcomplete dictionaries

Resolution in Magnetic Resonance (MR) is limited by diverse physical, technological and economical considerations. In conventional medical practice, resolution enhancement is usually performed with bicubic or B-spline interpolations, strongly affecting the accuracy of subsequent processing steps such as segmentation or registration. This paper presents a sparse-based super-resolution method, adapted for easily including prior knowledge, which couples up high and low frequency information so that a high-resolution version of a low-resolution brain MR image is generated. The proposed approach includes a whole-image multi-scale edge analysis and a dimensionality reduction scheme, which results in a remarkable improvement of the computational speed and accuracy, taking nearly 26 min to generate a complete 3D high-resolution reconstruction. The method was validated by comparing interpolated and reconstructed versions of 29 MR brain volumes with the original images, acquired in a 3T scanner, obtaining a reduction of 70% in the root mean squared error, an increment of 10.3 dB in the peak signal-to-noise ratio, and an agreement of 85% in the binary gray matter segmentations. The proposed method is shown to outperform a recent state-of-the-art algorithm, suggesting a substantial impact in voxel-based morphometry studies.

Content-based histopathology image retrieval using a kernel-based semantic annotation framework

Large amounts of histology images are captured and archived in pathology departments due to the ever expanding use of digital microscopy. The ability to manage and access these collections of digital images is regarded as a key component of next generation medical imaging systems. This paper addresses the problem of retrieving histopathology images from a large collection using an example image as query. The proposed approach automatically annotates the images in the collection, as well as the query images, with high-level semantic concepts. This semantic representation delivers an improved retrieval performance providing more meaningful results. We model the problem of automatic image annotation using kernel methods, resulting in a unified framework that includes: (1) multiple features for image representation, (2) a feature integration and selection mechanism (3) and an automatic semantic image annotation strategy. An extensive experimental evaluation demonstrated the effectiveness of the proposed framework to build meaningful image representations for learning and useful semantic annotations for image retrieval.

Topology-corrected segmentation and local intensity estimates for improved partial volume classification of brain cortex in MRI.

In magnetic resonance imaging (MRI), accuracy and precision with which brain structures may be quantified are frequently affected by the partial volume (PV) effect. PV is due to the limited spatial resolution of MRI compared to the size of anatomical structures. Accurate classification of mixed voxels and correct estimation of the proportion of each pure tissue (fractional content) may help to increase the precision of cortical thickness estimation in regions where this measure is particularly difficult, such as deep sulci. The contribution of this work is twofold: on the one hand, we propose a new method to label voxels and compute tissue fractional content, integrating a mechanism for detecting sulci with topology preserving operators. On the other hand, we improve the computation of the fractional content of mixed voxels using local estimation of pure tissue intensity means. Accuracy and precision were assessed using simulated and real MR data and comparison with other existing approaches demonstrated the benefits of our method. Significant improvements in gray matter (GM) classification and cortical thickness estimation were brought by the topology correction. The fractional content root mean squared error diminished by 6.3% (p<0.01) on simulated data. The reproducibility error decreased by 8.8% (p<0.001) and the Jaccard similarity measure increased by 3.5% on real data. Furthermore, compared with manually guided expert segmentations, the similarity measure was improved by 12.0% (p<0.001). Thickness estimation with the proposed method showed a higher reproducibility compared with the measure performed after partial volume classification using other methods.

An unsupervised feature learning framework for basal cell carcinoma image analysis

OBJECTIVE The paper addresses the problem of automatic detection of basal cell carcinoma (BCC) in histopathology images. In particular, it proposes a framework to both, learn the image representation in an unsupervised way and visualize discriminative features supported by the learned model. MATERIALS AND METHODS This paper presents an integrated unsupervised feature learning (UFL) framework for histopathology image analysis that comprises three main stages: (1) local (patch) representation learning using different strategies (sparse autoencoders, reconstruct independent component analysis and topographic independent component analysis (TICA), (2) global (image) representation learning using a bag-of-features representation or a convolutional neural network, and (3) a visual interpretation layer to highlight the most discriminant regions detected by the model. The integrated unsupervised feature learning framework was exhaustively evaluated in a histopathology image dataset for BCC diagnosis. RESULTS The experimental evaluation produced a classification performance of 98.1%, in terms of the area under receiver-operating-characteristic curve, for the proposed framework outperforming by 7% the state-of-the-art discrete cosine transform patch-based representation. CONCLUSIONS The proposed UFL-representation-based approach outperforms state-of-the-art methods for BCC detection. Thanks to its visual interpretation layer, the method is able to highlight discriminative tissue regions providing a better diagnosis support. Among the different UFL strategies tested, TICA-learned features exhibited the best performance thanks to its ability to capture low-level invariances, which are inherent to the nature of the problem.

A semantic content-based retrieval method for histopathology images

This paper proposes a model for content-based retrieval of histopathology images. The most remarkable characteristic of the proposed model is that it is able to extract high-level features that reflect the semantic content of the images. This is accomplished by a semantic mapper that maps conventional low-level features to high-level features using state-of-the-art machine-learning techniques. The semantic mapper is trained using images labeled by a pathologist. The system was tested on a collection of 1502 histopathology images and the performance assessed using standard measures. The results show an improvement from a 67% of average precision for the first result, using low-level features, to 80% of precision using high-level features.

Automatic annotation of histopathological images using a latent topic model based on non-negative matrix factorization

Histopathological images are an important resource for clinical diagnosis and biomedical research. From an image understanding point of view, the automatic annotation of these images is a challenging problem. This paper presents a new method for automatic histopathological image annotation based on three complementary strategies, first, a part-based image representation, called the bag of features, which takes advantage of the natural redundancy of histopathological images for capturing the fundamental patterns of biological structures, second, a latent topic model, based on non-negative matrix factorization, which captures the high-level visual patterns hidden in the image, and, third, a probabilistic annotation model that links visual appearance of morphological and architectural features associated to 10 histopathological image annotations. The method was evaluated using 1,604 annotated images of skin tissues, which included normal and pathological architectural and morphological features, obtaining a recall of 74% and a precision of 50%, which improved a baseline annotation method based on support vector machines in a 64% and 24%, respectively.

An experimental study of pathologist's navigation patterns in virtual microscopy

In virtual microscopy, a sequential process of captures of microscopical fields, allows to construct a virtual slide which is visualized using a specialized software, called the virtual microscopy viewer. This tool allows useful exploration of images, composed of thousands of microscopical fields of view at different levels of magnification, emulating an actual microscopical examination. The aim of this study was to establish the main pathologists navigation patterns when exploring virtual microscopy slides, using a graphical user interface, adapted to the pathologists workflow. Four pathologists with a similar level of experience, graduated from the same pathology program, navigated six virtual slides. Different issues were evaluated, namely, the percentage of common visited image regions, the time spent at each and its coincidence level, that is to say, the region of interest location. In addition, navigation patterns were also assessed, i.e., mouse movement velocities and linearity of the diagnostic paths. Results suggest that regions of interest are determined by a complex combination of the visited area, the time spent at each visit and the coincidence level among pathologists. Additionally, linear trajectories and particular velocity patterns were found for the registered diagnostic paths.

Infected cell identification in thin blood images based on color pixel classification: comparison and analysis

Malaria is an infectious disease which is mainly diagnosed by visual microscopical evaluation of Giemsa-stained thin blood films using a differential analysis of color features. This paper presents the evaluation of a color segmentation technique, based on standard supervised classification algorithms. The whole approach uses a general purpose classifier, which is parameterized and adapted to the problem of separating image pixels into three different classes: parasite, blood red cells and background. Assessment included not only four different supervised classification techniques - KNN, Naive Bayes, SVM and MLP - but different color spaces -RGB, normalized RGB, HSV and YCbCr-. Results show better performance for the KNN classifiers along with an improving feature characterization in the normalized RGB color space.

Segmentation of pelvic structures for planning CT using a geometrical shape model tuned by a multi-scale edge detector.

Accurate segmentation of the prostate and organs at risk in computed tomography (CT) images is a crucial step for radiotherapy planning. Manual segmentation, as performed nowadays, is a time consuming process and prone to errors due to the a high intra- and inter-expert variability. This paper introduces a new automatic method for prostate, rectum and bladder segmentation in planning CT using a geometrical shape model under a Bayesian framework. A set of prior organ shapes are first built by applying principal component analysis to a population of manually delineated CT images. Then, for a given individual, the most similar shape is obtained by mapping a set of multi-scale edge observations to the space of organs with a customized likelihood function. Finally, the selected shape is locally deformed to adjust the edges of each organ. Experiments were performed with real data from a population of 116 patients treated for prostate cancer. The data set was split in training and test groups, with 30 and 86 patients, respectively. Results show that the method produces competitive segmentations w.r.t standard methods (averaged dice = 0.91 for prostate, 0.94 for bladder, 0.89 for rectum) and outperforms the majority-vote multi-atlas approaches (using rigid registration, free-form deformation and the demons algorithm).