Comparative study of convolutional neural network architectures for gastrointestinal lesions classification.

The gastrointestinal (GI) tract can be affected by different diseases or lesions such as esophagitis, ulcers, hemorrhoids, and polyps, among others. Some of them can be precursors of cancer such as polyps. Endoscopy is the standard procedure for the detection of these lesions. The main drawback of this procedure is that the diagnosis depends on the expertise of the doctor. This means that some important findings may be missed. In recent years, this problem has been addressed by deep learning (DL) techniques. Endoscopic studies use digital images. The most widely used DL technique for image processing is the convolutional neural network (CNN) due to its high accuracy for modeling complex phenomena. There are different CNNs that are characterized by their architecture. In this article, four architectures are compared: AlexNet, DenseNet-201, Inception-v3, and ResNet-101. To determine which architecture best classifies GI tract lesions, a set of metrics; accuracy, precision, sensitivity, specificity, F1-score, and area under the curve (AUC) were used. These architectures were trained and tested on the HyperKvasir dataset. From this dataset, a total of 6,792 images corresponding to 10 findings were used. A transfer learning approach and a data augmentation technique were applied. The best performing architecture was DenseNet-201, whose results were: 97.11% of accuracy, 96.3% sensitivity, 99.67% specificity, and 95% AUC.

Radiogenomics analysis identifies correlations of digital mammography with clinical molecular signatures in breast cancer.

In breast cancer, well-known gene expression subtypes have been related to a specific clinical outcome. However, their impact on the breast tissue phenotype has been poorly studied. Here, we investigate the association of imaging data of tumors to gene expression signatures from 71 patients with breast cancer that underwent pre-treatment digital mammograms and tumor biopsies. From digital mammograms, a semi-automated radiogenomics analysis generated 1,078 features describing the shape, signal distribution, and texture of tumors along their contralateral image used as control. From tumor biopsy, we estimated the OncotypeDX and PAM50 recurrence scores using gene expression microarrays. Then, we used multivariate analysis under stringent cross-validation to train models predicting recurrence scores. Few univariate features reached Spearman correlation coefficients above 0.4. Nevertheless, multivariate analysis yielded significantly correlated models for both signatures (correlation of OncotypeDX = 0.49 ± 0.07 and PAM50 = 0.32 ± 0.10 in stringent cross-validation and OncotypeDX = 0.83 and PAM50 = 0.78 for a unique model). Equivalent models trained from the unaffected contralateral breast were not correlated suggesting that the image signatures were tumor-specific and that overfitting was not a considerable issue. We also noted that models were improved by combining clinical information (triple negative status and progesterone receptor). The models used mostly wavelets and fractal features suggesting their importance to capture tumor information. Our results suggest that molecular-based recurrence risk and breast cancer subtypes have observable radiographic phenotypes. To our knowledge, this is the first study associating mammographic information to gene expression recurrence signatures.

Nested case-control study reveals increased levels of urinary proteins from human kidney toxicity panels in women predicted to develop preeclampsia.

PURPOSE: The aim of this study was to evaluate the usefulness of urine concentrations of 12 proteins as a risk parameter for developing preeclampsia (PE). METHODS: A nested case-control study was designed to determine protein concentrations in urine from women predicted to develop PE (WPD-PE) and normotensive pregnancies (controls). Protein profiles were determined at 12, 16 and 20 gestational weeks (GW) using the Bio-Plex Pro human kidney toxicity Panel 1 and Panel 2 (Bio-Rad). Receiver operating characteristic (ROC) curve analyses were performed. Correlations between proteins and clinical parameters at the time of PE diagnosis were also assessed. RESULTS: Significant differences were observed in urine cystatin C (Cys C) levels at 16 and 20 GW and clusterin at 20 GW between WPD-PE and controls (P < 0.05). ROC analysis revealed that Cys C at 16 GW had the highest area under the ROC curve (0.758). At 16 GW, patients with urine Cys C levels above 73.7 ng/mL had eightfold increased odds for developing PE (odds ratio 7.92; 95 % CI 1.3-47.5; P = 0.027). A positive correlation was found between urinary Cys C (at 16 and 20 GW) and leukocyte counts, total proteins, aspartate aminotransferase, alanine aminotransferase, bilirubin and lactate dehydrogenase at the time of PE diagnosis (P value < 0.05). CONCLUSIONS: Urinary Cys C and clusterin showed predictive value for PE development in our cohort. Further studies are needed to validate their use as predictive biomarkers for PE and/or their participation in PE pathogenesis.