The present study aimed to develop an AI-based system for the detection and classification of polyps using colonoscopy images. A total of about 256,220 colonoscopy images from 5000 colorectal cancer patients were collected and processed. We used the CNN model for polyp detection and the EfficientNet-b0 model for polyp classification. Data were partitioned into training, validation and testing sets, with a 70%, 15% and 15% ratio, respectively. After the model was trained/validated/tested, to evaluate its performance rigorously, we conducted a further external validation using both prospective (n = 150) and retrospective (n = 385) approaches for data collection from 3 hospitals. The deep learning model performance with the testing set reached a state-of-the-art sensitivity and specificity of 0.9709 (95% CI: 0.9646–0.9757) and 0.9701 (95% CI: 0.9663–0.9749), respectively, for polyp detection. The polyp classification model attained an AUC of 0.9989 (95% CI: 0.9954–1.00). The external validation from 3 hospital results achieved 0.9516 (95% CI: 0.9295–0.9670) with the lesion-based sensitivity and a frame-based specificity of 0.9720 (95% CI: 0.9713–0.9726) for polyp detection. The model achieved an AUC of 0.9521 (95% CI: 0.9308–0.9734) for polyp classification. The high-performance, deep-learning-based system could be used in clinical practice to facilitate rapid, efficient and reliable decisions by physicians and endoscopists.

Breast cancer is a heterogeneously complex disease. A number of molecular subtypes with distinct biological features lead to different treatment responses and clinical outcomes. Traditionally, breast cancer is classified into subtypes based on gene expression profiles; these subtypes include luminal A, luminal B, basal like, HER2-enriched, and normal-like breast cancer. This molecular taxonomy, however, could only be appraised through transcriptome analyses. Our study applies deep convolutional neural networks and transfer learning from three pre-trained models, namely ResNet50, InceptionV3 and VGG16, for classifying molecular subtypes of breast cancer using TCGA-BRCA dataset. We used 20 whole slide pathological images for each breast cancer subtype. The results showed that our scale training reached about 78% of accuracy for validation. This outcomes suggested that classification of molecular subtypes of breast cancer by pathological images are feasible and could provide reliable results

This study aimed to conduct functional proteomics across breast cancer subtypes with bioinformatics analyses. Candidate proteins were identified using nanoscale liquid chromatography with tandem mass spectrometry (NanoLC-MS/MS) from core needle biopsy samples of early stage (0-III) breast cancers, followed by external validation with public domain gene-expression datasets (TCGA TARGET GTEx and TCGA BRCA). Seventeen proteins demonstrated significantly differential expression and protein-protein interaction (PPI) found the strong networks including COL2A1, COL11A1, COL6A1, COL6A2, THBS1 and LUM. Public domain databases also showed that COL2A1, COL11A1, COL6A1, COL6A2 and LUM were higher in primary/metastatic tumor than in normal tissue (one-way ANOVA, all P-values less than 0.001), and all six genes were differentially expressed across four molecular subtypes based on hormone receptor (HR) status and human epidermal growth factor receptor II (HER2) status (one-way ANOVA, all P-values less than 0.001). Disease-specific survival discrepancy was observed comparing breast cancer patients of the upper and lower quartile of the collagen family (COL2A1, COL11A1, COL6A1, COL6A2), THBS1 and LUM gene expression signature (log-rank test, P = 0.06). Functional proteomics suggested that collagen proteins, thrombospondin 1 and lumican are differentially expressed across breast cancer subtypes.