Germán Corredor

Feature Driven Local Cell Graph (FeDeG): Predicting Overall Survival in Early Stage Lung Cancer

The local spatial arrangement of nuclei in histopathology image has been shown to have prognostic value in the context of different cancers. In order to capture the nuclear architectural information locally, local cell cluster graph based measurements have been proposed. However, conventional ways of cell graph construction only utilize nuclear spatial proximity, and do not differentiate different cell types while constructing a cell graph. In this paper, we present feature driven local cell graph (FeDeG), a new approach to constructing local cell graphs by simultaneously considering spatial proximity and attributes of the individual nuclei (e.g. shape, size, texture). In addition, we designed a new set of quantitative graph derived metrics to be extracted from FeDeGs, in turn capturing the interplay between different local cell clusters. We evaluated the efficacy of FeDeG features in a digitized H&E stained tissue micro-array (TMA) images cohort consists of 434 early stage non-small cell lung cancer for predicting short-term ( 5 years) survival. Across a 100 runs of 10-fold cross-validation, a linear discriminant classifier in conjunction with the 15 most predictive FeDeG features identified via the Wilcoxon Rank Sum Test (WRST) yielded an average of AUC = 0.68. By comparison, four state-of-the-art pathomic and a deep learning based classifier had a corresponding AUC of 0.56, 0.54, 0.61, 0.62, and 0.55 respectively.

A watershed and feature-based approach for automated detection of lymphocytes on lung cancer images.

Automatic detection of lymphocytes could contribute to develop objective measures of the infiltration grade of tumors, which can be used by pathologists for improving the decision making and treatment planning processes. In this article, a simple framework to automatically detect lymphocytes on lung cancer images is presented. This approach starts by automatically segmenting nuclei using a watershed-based approach. Nuclei shape, texture, and color features are then used to classify each candidate nucleus as either lymphocyte or non-lymphocyte by a trained SVM classifier. Validation was carried out using a dataset containing 3420 annotated structures (lymphocytes and non-lymphocytes) from 13 1000 × 1000 fields of view extracted from lung cancer whole slide images. A Deep Learning model was trained as a baseline. Results show an F-score 30% higher with the presented framework than with the Deep Learning approach. The presented strategy is, in addition, more flexible, requires less computational power, and requires much lower training times.

Training a cell-level classifier for detecting basal-cell carcinoma by combining human visual attention maps with low-level handcrafted features

Abstract. Computational histomorphometric approaches typically use low-level image features for building machine learning classifiers. However, these approaches usually ignore high-level expert knowledge. A computational model (M_im) combines low-, mid-, and high-level image information to predict the likelihood of cancer in whole slide images. Handcrafted low- and mid-level features are computed from area, color, and spatial nuclei distributions. High-level information is implicitly captured from the recorded navigations of pathologists while exploring whole slide images during diagnostic tasks. This model was validated by predicting the presence of cancer in a set of unseen fields of view. The available database was composed of 24 cases of basal-cell carcinoma, from which 17 served to estimate the model parameters and the remaining 7 comprised the evaluation set. A total of 274 fields of view of size 1024×1024  pixels were extracted from the evaluation set. Then 176 patches from this set were used to train a support vector machine classifier to predict the presence of cancer on a patch-by-patch basis while the remaining 98 image patches were used for independent testing, ensuring that the training and test sets do not comprise patches from the same patient. A baseline model (M_ex) estimated the cancer likelihood for each of the image patches. M_ex uses the same visual features as M_im, but its weights are estimated from nuclei manually labeled as cancerous or noncancerous by a pathologist. M_im achieved an accuracy of 74.49% and an F-measure of 80.31%, while M_ex yielded corresponding accuracy and F-measures of 73.47% and 77.97%, respectively.

Multi-channel non-invasive fetal electrocardiography detection using wavelet decomposition.

Non-invasive fetal electrocardiography (fECG) has attracted the medical community because of the importance of fetal monitoring. However, its implementation in clinical practice is challenging: the fetal signal has a low Signal- to-Noise-Ratio and several signal sources are present in the maternal abdominal electrocardiography (AECG). This paper presents a novel method to detect the fetal signal from a multi-channel maternal AECG. The method begins by applying filters and signal detrending the AECG signals. Afterwards, the maternal QRS complexes are identified and subtracted. The residual signals are used to detect the fetal QRS complex. Intervals of these signals are analyzed by using a wavelet decomposition. The resulting representation feds a previously trained Random Forest (RF) classifier that identifies signal intervals associated to fetal QRS complex. The method was evaluated on a public available dataset: the Physionet2013 challenge. A set of 50 maternal AECG records were used to train the RF classifier. The evaluation was carried out in signals intervals extracted from additional 25 maternal AECG. The proposed method yielded an 83:77% accuracy in the fetal QRS complex classification task.

A sparse representation of the pathologist's interaction with whole slide images to improve the assigned relevance of regions of interest.

During a diagnosis task, a Pathologist looks over a Whole Slide Image (WSI), aiming to find out relevant pathological patterns. Nonetheless, a virtual microscope captures these structures, but also other cellular patterns with different or none diagnostic meaning. Annotation of these images depends on manual delineation, which in practice becomes a hard task. This article contributes a new method for detecting relevant regions in WSI using the routine navigations in a virtual microscope. This method constructs a sparse representation or dictionary of each navigation path and determines the hidden relevance by maximizing the incoherence between several paths. The resulting dictionaries are then projected onto each other and relevant information is set to the dictionary atoms whose similarity is higher than a custom threshold. Evaluation was performed with 6 pathological images segmented from a skin biopsy already diagnosed with basal cell carcinoma (BCC). Results show that our proposal outperforms the baseline by more than 20%.

A lymphocyte spatial distribution graph-based method for automated classification of recurrence risk on lung cancer images

Tumor-infiltrating lymphocytes occurs when various classes of white blood cells migrate from the blood stream towards the tumor, infiltrating it. The presence of TIL is predictive of the response of the patient to therapy. In this paper, we show how the automatic detection of lymphocytes in digital H and E histopathological images and the quantitative evaluation of the global lymphocyte configuration, evaluated through global features extracted from non-parametric graphs, constructed from the lymphocytes’ detected positions, can be correlated to the patient’s outcome in early-stage non-small cell lung cancer (NSCLC). The method was assessed on a tissue microarray cohort composed of 63 NSCLC cases. From the evaluated graphs, minimum spanning trees and K-nn showed the highest predictive ability, yielding F1 Scores of 0.75 and 0.72 and accuracies of 0.67 and 0.69, respectively. The predictive power of the proposed methodology indicates that graphs may be used to develop objective measures of the infiltration grade of tumors, which can, in turn, be used by pathologists to improve the decision making and treatment planning processes.

Quantifying expert diagnosis variability when grading tumor-infiltrating lymphocytes

Tumor-infiltrating lymphocytes (TILs) have proved to play an important role in predicting prognosis, survival, and response to treatment in patients with a variety of solid tumors. Unfortunately, currently, there are not a standardized methodology to quantify the infiltration grade. The aim of this work is to evaluate variability among the reports of TILs given by a group of pathologists who examined a set of digitized Non-Small Cell Lung Cancer samples (n=60). 28 pathologists answered a different number of histopathological images. The agreement among pathologists was evaluated by computing the Kappa index coefficient and the standard deviation of their estimations. Furthermore, TILs reports were correlated with patient’s prognosis and survival using the Pearson’s correlation coefficient. General results show that the agreement among experts grading TILs in the dataset is low since Kappa values remain below 0.4 and the standard deviation values demonstrate that in none of the images there was a full consensus. Finally, the correlation coefficient for each pathologist also reveals a low association between the pathologists’ predictions and the prognosis/survival data. Results suggest the need of defining standardized, objective, and effective strategies to evaluate TILs, so they could be used as a biomarker in the daily routine.

A radiology image retrieval system based on user preferences

This article introduces a computer-aided solution for radiology education integrated with the clinic practice, inherited from modern technologies that facilitate the processing of large amounts of stored information. Radiology training may have several challenges such as image retrieval, extraction of knowledge, education towards solving problems and the interaction with huge repositories known as Picture Archiving and Communication Systems (PACS). This project proposes a user-based system that learns from user interaction, retrieving not just the requested information but recommending related cases and interesting images. The recommended images are retrieved using a Click-through rate (CTR) strategy for defining the most similar cases in the database. This is a fully web-based proposal, potentially useful at classroom or home, that allows students to develop the clinical skills needed in a more realistic scenario.

Nuclei graph local features for basal cell carcinoma classification in whole slide images

Evidence based medicine aims to provide a quantifiable framework to support cancer optimal treatment selection. Pathological examination is the main evidence used in medical management, yet the level of quantification is low and highly dependent on the examiner expertise. This paper presents and evaluates a method to extract graph based topological features from skin tissue images to identify cancerous regions associated to basal cell carcinoma. The graph features constitute a quantitative measure of the architectural tissue organization. Results show that graph topological features extracted from a nuclei based distance graph, particularly those related to local density, have a high predictive value in the automated detection of basal cell carcinoma. The method was evaluated using a leave-one-out validation scheme in a set of 9 skin Whole Slide Images obtaining a 0.76 F-score in distinguishing basal cell carcinoma regions in skin tissue whole slide images.

Registration of cardiac magnetic resonance images using SURF points and matching

Reconstruction of the heartbeat is an useful tool to detect and diagnose some pathologies. However, this process represents a challenge because the heart is a moving organ inside a moving body, so that, either similar regions are hard to identify or some regions appear and disappear constantly. This article presents a reconstruction method of the right ventricle using SURF points in irregular regions. The SURF points, invariant to image scale and rotation, provide robust features of a right ventricle slice that can then be traced to the other slices. By using such points and then, using a nonrigid registration, it possible to perform a volumetrical reconstruction of these images.