Filipe Janela

3D Lacunarity in Multifractal Analysis of Breast Tumor Lesions in Dynamic Contrast-Enhanced Magnetic Resonance Imaging

Dynamic contrast-enhanced magnetic resonance (DCE-MR) of the breast is especially robust for the diagnosis of cancer in high-risk women due to its high sensitivity. Its specificity may be, however, compromised since several benign masses take up contrast agent as malignant lesions do. In this paper, we propose a novel method of 3D multifractal analysis to characterize the spatial complexity (spatial arrangement of texture) of breast tumors at multiple scales. Self-similar properties are extracted from the estimation of the multifractal scaling exponent for each clinical case, using lacunarity as the multifractal measure. These properties include several descriptors of the multifractal spectra reflecting the morphology and internal spatial structure of the enhanced lesions relatively to normal tissue. The results suggest that the combined multifractal characteristics can be effective to distinguish benign and malignant findings, judged by the performance of the support vector machine classification method evaluated by receiver operating characteristics with an area under the curve of 0.96. In addition, this paper confirms the presence of multifractality in DCE-MR volumes of the breast, whereby multiple degrees of self-similarity prevail at multiple scales. The proposed feature extraction and classification method have the potential to complement the interpretation of the radiologists and supply a computer-aided diagnosis system.

Towards the detection of microcalcifications on mammograms through Multifractal Detrended Fluctuation Analysis

In this paper, we propose a method based on a generalization of the multifractal detrended fluctuation analysis (MF-DFA) to the two-dimensionality, for the analysis of breast medical images, particularly grey scale images of mammograms. The existent generalization has been suitably applied in synthetic multifractal surfaces. However, when it is applied to natural images only some scaling laws are revealed, not reinforcing the relevance of the method in the field. Therefore, it is shown that the proposed method is appropriate for detecting signs of breast cancer in mammograms, such as microcalcifications, based on recent approaches of the self-similarity formalism. The accuracy of the method is analysed and discussed when applied to mammograms, revealing features distinguished by direct singularity information extraction.

Classification of Breast Masses on Contrast-Enhanced Magnetic Resonance Images Through Log Detrended Fluctuation Cumulant-Based Multifractal Analysis

This paper proposes a multiscale automated model for the classification of suspicious malignancy of breast masses, through log detrended fluctuation cumulant-based multifractal analysis of images acquired by dynamic contrast-enhanced magnetic resonance. Features for classification are extracted by computing the multifractal scaling exponent for each of the 70 clinical cases and by quantifying the log-cumulants reflecting multifractal information related with texture of the enhanced lesions. The output is compared with the radiologist diagnosis that follows the Breast Imaging-Reporting and Data System (BI-RADS). The results suggest that the log-cumulant C2 can be effective in classifying typically biopsy-recommended cases. The performance of a supervised classification was evaluated by receiver operating characteristic (ROC) with an area under the curve of 0.985. The proposed multifractal analysis can contribute to novel feature classification techniques to aid radiologists every time there is a change in the clinical course, namely, when biopsy should be considered.

Using Structured EHR Data and SVM to Support ICD-9-CM Coding

This study proposes a methodology to support coding professionals in assigning ICD-9-CM codes to inpatient episodes. This subject has been predominantly addressed through the use of natural language processing methods, which show limited generalizability. To surpass this issue, this paper proposes a methodology entailing an adaptive data processing method based on structured electronic health record data, whereby raw clinical data is mapped into a feature set, and based on which supervised learning algorithms are trained. After applying a filter method for feature selection, support vector machine (SVM) classifiers are trained to obtain predictions for assigning codes to each episode. This approach is tested using a dataset of inpatient episodes from a department of Internal Medicine. Classifiers exhibited F1-measure values around 52%. Recall was generally higher than precision, which is considered valuable for coding support purposes. Analyzing results on an individual code basis sheds light on some key-issues regarding the use of structured electronic health record data in supporting clinical coding.

Optimisation-Based on Simulation: A Diagnostic Imaging Department Case-Study

The growing influx of patients in healthcare providers is the result of an aging population and emerging self-consciousness about health. Thus, it is necessary to implement methodologies that optimize the healthcare providers efficiency while increasing patient throughput and reducing patients total waiting time. This paper presents a case study of a conventional radiology workflow analysis in a Portuguese Healthcare Provider. Modeling tools were applied to define the existing workflow. Reengineered workflows were analyzed in a simulation tool. The integration of modeling and simulation tools allowed the identification of system bottlenecks. This study is focused in an Imaging Department and resulted in the identification of a new workflow that permits the relative reduction of the total completion time in 41%.

Clinical coding support based on structured data stored in electronic health records

Clinical coding is an increasingly essential process within health organizations, usually performed manually and entailing several challenges: its administrative burden, raising costs and eventual errors. To address this issue, several coding support systems have been proposed across the literature. However, these systems are based on text processing methods that may be limited by poor text quality, ambiguity and lack of annotated resources. As electronic health record systems tend to implement more structured data formats, we propose a methodology for coding support based on structured clinical data collected during inpatient care from a semi-structured electronic health record. We follow a statistical learning paradigm and investigate several building blocks of the methodology to assess the feasibility of the approach. We present and discuss preliminary results obtained with real data extracted from an Internal Medicine department and identify several measures to further develop the methodology, model performance and generalizability.

Multifractal analysis of Arterial Spin Labeling functional Magnetic Resonance Imaging of the brain

This work proposes a multifractal analysis of the time series derived from ASL fMRI (Arterial Spin Labeling functional Magnetic Resonance Imaging) to detect brain activated regions in response to an unknown stimulus. In contrast to standard model-based activation analysis, no prior knowledge of the expected haemodynamic response has to be assumed for extracting activation patterns from fMRI. The ASL time series were analysed using MF-DFA (Multifractal Detrended Fluctuation Analysis). The results show clear differences between the multifractal spectra, in form and locus, with respect to voxels from activated and non-activated brain regions. These results are in line with known literature for BOLD (Blood Oxygenation Level Dependent) functional time series. MF-DFA reveals stronger activation in the motor cortex and in some other physiologically relevant activation areas, as the visual cortex. It is shown that the proposed framework is appropriate for studying the human brain function, based on recent approaches of the self-similarity formalism.

Preprocessing structured clinical data for predictive modeling and decision support. A roadmap to tackle the challenges.

BACKGROUND EHR systems have high potential to improve healthcare delivery and management. Although structured EHR data generates information in machine-readable formats, their use for decision support still poses technical challenges for researchers due to the need to preprocess and convert data into a matrix format. During our research, we observed that clinical informatics literature does not provide guidance for researchers on how to build this matrix while avoiding potential pitfalls. OBJECTIVES This article aims to provide researchers a roadmap of the main technical challenges of preprocessing structured EHR data and possible strategies to overcome them. METHODS Along standard data processing stages - extracting database entries, defining features, processing data, assessing feature values and integrating data elements, within an EDPAI framework -, we identified the main challenges faced by researchers and reflect on how to address those challenges based on lessons learned from our research experience and on best practices from related literature. We highlight the main potential sources of error, present strategies to approach those challenges and discuss implications of these strategies. RESULTS Following the EDPAI framework, researchers face five key challenges: (1) gathering and integrating data, (2) identifying and handling different feature types, (3) combining features to handle redundancy and granularity, (4) addressing data missingness, and (5) handling multiple feature values. Strategies to address these challenges include: cross-checking identifiers for robust data retrieval and integration; applying clinical knowledge in identifying feature types, in addressing redundancy and granularity, and in accommodating multiple feature values; and investigating missing patterns adequately. CONCLUSIONS This article contributes to literature by providing a roadmap to inform structured EHR data preprocessing. It may advise researchers on potential pitfalls and implications of methodological decisions in handling structured data, so as to avoid biases and help realize the benefits of the secondary use of EHR data.

GATE Mammogram Simulation of NCAT Breast Phantom

In this paper, we aim to acquire simulated mammograms. This simulated data is performed on GATE (Geant4 Application for Tomographic Emission), a platform for simulation based on Monte Carlo methods. On this software, a particular tool, the new CTscanner, is used to simulate the interation between X-ray and tissues. We hope also to demonstrate how the new CTscanner system can be used. The NCAT phantom is used to obtain data, where only the volume of the breast is employed to achieve the simulation. Geometric phantoms are also used for simulation of more simple objects. The parameters for the reproduction of mammograms are introduced accordingly the Siemens Mammomat Inspiration mammography unit, since this is a recent device. The aim of this work is to get simulated mammograms from different angles of the same object, which will allow the access of tri-dimensional reconstruction for Digital Breast Tomosynthesis. We present and discuss the results obtained using the proposed simulations.

Pipeline for spatial distortion correction in MRI

The recent advances in Magnetic Resonance Imaging gradient, regarding strength and computation speed, led to the development of Echo-Planar Imaging pulse-sequences with faster acquisition times. This kind of sequence is used in functional MRI and diffusion-weighted Magnetic Resonance Imaging and it presents more distortions than slower sequences. This work aims to compare different spatial distortion correction methods for Echo-Planar Imaging sequences with a new proposed pipeline which consists in performing a Field Map correction after a registration process.