AI in cancer detection

This paper uses weakly supervised contrastive learning to infer molecular subtypes and recurrence risk in breast cancer from unannotated pathology images, achieving strong cross-cohort performance and alignment with pathologist annotations.

This paper compares single and ensemble CNN models for detecting invasive breast cancer in histological images, finding ensembles improve accuracy across internal and external datasets.

This paper introduces wsi2rppa, a weakly supervised contrastive learning framework that infers protein levels and drug response in breast cancer from whole slide images, achieving 0.79 AUC for trastuzumab response prediction.

This study uses attention-based multiple instance learning to predict HRD status from H&E images across ten tumor types, achieving moderate to high AUROC values.

This study develops an R-based method to accurately measure DAB staining on immunohistochemical slides, achieving a mean absolute error of 0.76% compared to manual measurements.

This study develops a deep learning model to classify head and neck squamous lesions with confidence scores to identify uncertain predictions.

This paper introduces PGDIPS, a physics-guided deep learning method for stain deconvolution in histopathology images, achieving better generalization and interpretability without training data.

HAPPY quantifies cell and tissue variability in placenta histology images, providing metrics that align with clinical experts and reveal deviations in placental infarction.

The paper presents a web/cloud-based digital pathology framework for AI development and deployment, supporting collaborative annotation and remote slide access.

The paper presents a pan-cancer PDX histology image repository with genomic and pathological annotations for deep learning analysis of tumor regions and xenograft-transplant lymphoproliferative disorder prediction.

This paper applies a convolutional neural network to dynamic optical contrast imaging for tumor classification in head and neck cancers, achieving high accuracy in non-tumor tissue classification.

stainlib provides a Python library for augmenting and normalizing H&E histopathology images to improve model robustness across clinical settings.

This paper develops a CNN-based method to predict HER2 status in esophageal cancer tissue microarrays with high accuracy, outperforming manual assessment and reducing the need for costly follow-up tests.

This paper develops a deep learning approach for classifying lung adenocarcinoma patterns and non-small cell lung cancer types using histopathological images, achieving high accuracy in classification tasks.

This paper develops a deep learning model that accurately segments tumors using only eight whole-slide images, showing that small, well-annotated datasets can achieve high performance in medical imaging.

The paper proposes a modified cross-validation approach and a new neural network architecture for treatment response prediction in small Triple Negative Breast Cancer cohorts.

This paper develops a deep learning pipeline to predict mutation status from histopathological images in cancer patients, identifying clinically relevant genes like TP53 and proposing screening strategies.

The paper presents an automated pipeline for detecting and grading canine brain glioma using GFAP immunohistological images, validated through stereological indexes and machine learning.

This paper introduces DIME, a self-supervised learning pipeline for training image patch embeddings from whole slide images, enabling effective downstream tasks with limited data.

This paper introduces 'mones' as interpretable CNN features from H&E images that correlate with biological phenotypes and can accurately classify tumors and normals.

This study benchmarks AI methods for end-to-end computational pathology, showing ResNet and ViT outperform MIL and CLAM in mutation prediction but not in tumor subtyping.

The paper introduces DeepLIIF, a deep learning framework that translates IHC images to multiplex IF images and quantifies single-cell IHC markers.

This paper uses deep learning to analyze histopathology images of colorectal cancer to predict progression-free survival by quantifying tumor infiltrating lymphocytes.

The paper does not provide substantive content as the authors have withdrawn the manuscript due to data-sharing issues.

This study shows that transfer learning from synthetic OCT image simulations improves classification of esophageal tissue in BE patients compared to models trained on clinical data with uncertain labels.

This paper develops a method to predict molecular phenotypes from cancer pathology images using interpretable features, showing correlations with tumor microenvironment markers and predictive power for molecular signatures.

This study uses GANs to synthesize high-resolution cancer pathology images, showing they are visually indistinguishable from real images and perform as well as real images in diagnostic tasks.

The paper uses deep convolutional neural networks on pathological images from the TCGA-BRCA dataset to predict breast cancer molecular subtypes with 78% validation accuracy.

This paper proposes a simple computer vision algorithm to assist pathologists in screening histological images for malignancy, focusing on high-risk areas to improve efficiency and standardization.