A pathomic approach for tumor-infiltrating lymphocytes classification on breast cancer digital pathology images

Background and objectives

The detection of tumor-infiltrating lymphocytes (TILs) could aid in the development of objective measures of the infiltration grade and can support decision-making in breast cancer (BC). However, manual quantification of TILs in BC histopathological whole slide images (WSI) is currently based on a visual assessment, thus resulting not standardized, not reproducible, and time-consuming for pathologists. In this work, a novel pathomic approach, aimed to apply high-throughput image feature extraction techniques to analyze the microscopic patterns in WSI, is proposed. In fact, pathomic features provide additional information concerning the underlying biological processes compared to the WSI visual interpretation, thus providing more easily interpretable and explainable results than the most frequently investigated Deep Learning based methods in the literature.

Methods

A dataset containing 1037 regions of interest with tissue compartments and TILs annotated on 195 TNBC and HER2+ BC hematoxylin and eosin (H&E)-stained WSI was used. After segmenting nuclei within tumor-associated stroma using a watershed-based approach, 71 pathomic features were extracted from each nucleus and reduced using a Spearman's correlation filter followed by a nonparametric Wilcoxon rank-sum test and least absolute shrinkage and selection operator. The relevant features were used to classify each candidate nucleus as either TILs or non-TILs using 5 multivariable machine learning classification models trained using 5-fold cross-validation (1) without resampling, (2) with the synthetic minority over-sampling technique and (3) with downsampling. The prediction performance of the models was assessed using ROC curves.

Results

21 features were selected, with most of them related to the well-known TILs properties of having regular shape, clearer margins, high peak intensity, more homogeneous enhancement and different textural pattern than other cells. The best performance was obtained by Random-Forest with ROC AUC of 0.86, regardless of resampling technique.

Conclusions

The presented approach holds promise for the classification of TILs in BC H&E-stained WSI and could provide support to pathologists for a reliable, rapid and interpretable clinical assessment of TILs in BC.

The relationship between radiomics and pathomics in Glioblastoma patients: Preliminary results from a cross-scale association study

Glioblastoma multiforme (GBM) typically exhibits substantial intratumoral heterogeneity at both microscopic and radiological resolution scales. Diffusion Weighted Imaging (DWI) and dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) are two functional MRI techniques that are commonly employed in clinic for the assessment of GBM tumor characteristics. This work presents initial results aiming at determining if radiomics features extracted from preoperative ADC maps and post-contrast T1 (T1C) images are associated with pathomic features arising from H&E digitized pathology images. 48 patients from the public available CPTAC-GBM database, for which both radiology and pathology images were available, were involved in the study. 91 radiomics features were extracted from ADC maps and post-contrast T1 images using PyRadiomics. 65 pathomic features were extracted from cell detection measurements from H&E images. Moreover, 91 features were extracted from cell density maps of H&E images at four different resolutions. Radiopathomic associations were evaluated by means of Spearman's correlation (ρ) and factor analysis. p values were adjusted for multiple correlations by using a false discovery rate adjustment. Significant cross-scale associations were identified between pathomics and ADC, both considering features (n = 186, 0.45 < ρ < 0.74 in absolute value) and factors (n = 5, 0.48 < ρ < 0.54 in absolute value). Significant but fewer ρ values were found concerning the association between pathomics and radiomics features (n = 53, 0.5 < ρ < 0.65 in absolute value) and factors (n = 2, ρ = 0.63 and ρ = 0.53 in absolute value). The results of this study suggest that cross-scale associations may exist between digital pathology and ADC and T1C imaging. This can be useful not only to improve the knowledge concerning GBM intratumoral heterogeneity, but also to strengthen the role of radiomics approach and its validation in clinical practice as "virtual biopsy", introducing new insights for omics integration toward a personalized medicine approach.

Middle-Level Features for the Explanation of Classification Systems by Sparse Dictionary Methods

Machine learning (ML) systems are affected by a pervasive lack of transparency. The eXplainable Artificial Intelligence (XAI) research area addresses this problem and the related issue of explaining the behavior of ML systems in terms that are understandable to human beings. In many explanation of XAI approaches, the output of ML systems are explained in terms of low-level features of their inputs. However, these approaches leave a substantive explanatory burden with human users, insofar as the latter are required to map low-level properties into more salient and readily understandable parts of the input. To alleviate this cognitive burden, an alternative model-agnostic framework is proposed here. This framework is instantiated to address explanation problems in the context of ML image classification systems, without relying on pixel relevance maps and other low-level features of the input. More specifically, one obtains sets of middle-level properties of classification inputs that are perceptually salient by applying sparse dictionary learning techniques. These middle-level properties are used as building blocks for explanations of image classifications. The achieved explanations are parsimonious, for their reliance on a limited set of middle-level image properties. And they can be contrastive, because the set of middle-level image properties can be used to explain why the system advanced the proposed classification over other antagonist classifications. In view of its model-agnostic character, the proposed framework is adaptable to a variety of other ML systems and explanation problems.

Unsupervised entity and relation extraction from clinical records in Italian

This paper proposes and discusses the use of text mining techniques for the extraction of information from clinical records written in Italian. However, as it is very difficult and expensive to obtain annotated material for languages different from English, we only consider unsupervised approaches, where no annotated training set is necessary. We therefore propose a complete system that is structured in two steps. In the first one domain entities are extracted from the clinical records by means of a metathesaurus and standard natural language processing tools. The second step attempts to discover relations between the entity pairs extracted from the whole set of clinical records. For this last step we investigate the performance of unsupervised methods such as clustering in the space of entity pairs, represented by an ad hoc feature vector. The resulting clusters are then automatically labelled by using the most significant features. The system has been tested on a fairly large data set of clinical records in Italian, investigating the variation in the performance adopting different similarity measures in the feature space. The results of our experiments show that the unsupervised approach proposed is promising and well suited for a semi-automatic labelling of the extracted relations.

Unsupervised information extraction from italian clinical records

This paper discusses the application of an unsupervised text mining technique for the extraction of information from clinical records in Italian. The approach includes two steps. First of all, a metathesaurus is exploited together with natural language processing tools to extract the domain entities. Then, clustering is applied to explore relations between entity pairs. The results of a preliminary experiment, performed on the text extracted from 57 medical records containing more than 20,000 potential relations, show how the clustering should be based on the cosine similarity distance rather than the City Block or Hamming ones.