Search

Background: For virtually every patient with colorectal cancer (CRC), hematoxylin-eosin (HE)-stained tissue slides are available. These images contain quantitative information, which is not routinely used to objectively extract prognostic biomarkers. In the present study, we investigated whether deep convolutional neural networks (CNNs) can extract prognosticators directly from these widely available images.Methods and Findings: We hand-delineated single-tissue regions in 86 CRC tissue slides, yielding more than 100,000 HE image patches, and used these to train a CNN by transfer learning, reaching a nine-class accuracy of >94% in an independent data set of 7,180 images from 25 CRC patients. With this tool, we performed automated tissue decomposition of representative multitissue HE images from 862 HE slides in 500 stage I-IV CRC patients in the The Cancer Genome Atlas (TCGA) cohort, a large international multicenter collection of CRC tissue. Based on the output neuron activations in the CNN, we calculated a "deep stroma score," which was an independent prognostic factor for overall survival (OS) in a multivariable Cox proportional hazard model (hazard ratio [HR] with 95% confidence interval [CI]: 1.99 [1.27-3.12], p = 0.0028), while in the same cohort, manual quantification of stromal areas and a gene expression signature of cancer-associated fibroblasts (CAFs) were only prognostic in specific tumor stages. We validated these findings in an independent cohort of 409 stage I-IV CRC patients from the "Darmkrebs: Chancen der Verhütung durch Screening" (DACHS) study who were recruited between 2003 and 2007 in multiple institutions in Germany. Again, the score was an independent prognostic factor for OS (HR 1.63 [1.14-2.33], p = 0.008), CRC-specific OS (HR 2.29 [1.5-3.48], p = 0.0004), and relapse-free survival (RFS; HR 1.92 [1.34-2.76], p = 0.0004). A prospective validation is required before this biomarker can be implemented in clinical workflows.Conclusions: In our retrospective study, we show that a CNN can assess the human tumor microenvironment and predict prognosis directly from histopathological images. [ABSTRACT FROM AUTHOR]

Imaging spectroscopy integrates conventional imaging and spectroscopy to attain both spatial and spectral information from an object. Processing and analysis of a hypercube can be a hard task. Robust methods for hyperspectral data classification are required, insensitive to imaging deficiencies and high input dimension. Mushrooms have a thin and porous epidermal structure, and are sensitive to handling and transportation practices. Mechanical damage triggers a browning process within the tissue changing its metabolic state. The objective is to quantify different levels of physical perturbation on the mushroom pilei, using near-infrared spectral images and machine learning approaches. An ANN classifier is implemented, whose input is a small set of vectors containing representative information, and output is the set of categorical labels that correspond to different levels of mechanical vibration. For obtaining a salient dataset for classifying the images, the Harris corner detection algorithm is employed. The advantage of using interest points is to replace an exhaustive search over the entire image space by a computation over a concise set of highly informative points. A frame-based classification approach is proposed and shown to produce an increase in the classification accuracy, since feature vectors regarded as single instances may not always carry sufficient discriminant information. Comparisons with statistical features computed from wavelet coefficients showed that interest points are more suitable in assessing mechanical perturbation. Comparisons on a classifier level with support vector machines showed that ANNs perform better for the specific application, implying a connection between the classification method and the underlying learning problem. Overall, the frame-based classification scheme reduced the misclassification rate. This approach is suited for challenging classification problems where the degree of class separation is variable, i.e., assessment of mechanical damage in mushrooms. [ABSTRACT FROM AUTHOR]

Abstract: Lacunarity is about quantifying the degree of spatial heterogeneity in the visual texture of imagery through the identification of the relationships between patterns and their spatial configurations in a two-dimensional setting. The computed lacunarity data can designate a mathematical index of spatial heterogeneity, therefore the corresponding feature vectors should possess the necessary inter-class statistical properties that would enable them to be used for pattern recognition purposes. The objectives of this study is to construct a supervised parsimonious classification model of binary lacunarity data–computed by Valous et al. (2009)–from pork ham slice surface images, with the aid of kernel principal component analysis (KPCA) and artificial neural networks (ANNs), using a portion of informative salient features. At first, the dimension of the initial space (510 features) was reduced by 90% in order to avoid any noise effects in the subsequent classification. Then, using KPCA, the first nineteen kernel principal components (99.04% of total variance) were extracted from the reduced feature space, and were used as input in the ANN. An adaptive feedforward multilayer perceptron (MLP) classifier was employed to obtain a suitable mapping from the input dataset. The correct classification percentages for the training, test and validation sets were 86.7%, 86.7%, and 85.0%, respectively. The results confirm that the classification performance was satisfactory. The binary lacunarity spatial metric captured relevant information that provided a good level of differentiation among pork ham slice images. [ABSTRACT FROM AUTHOR]