AI in cancer detection

The paper develops TOPAZ, a tool that predicts cancerous cells/glands and analyzes immune cell spatial relationships in prostate cancer, finding that immune cell distance from cancer glands correlates with recurrence risk.

Star-Motifs identifies spatial patterns in histology slides to reveal clinically relevant tumor microenvironment features and survival correlations.

AnnotateAnyCell reduces annotation time for histopathological images by 25% using a semi-supervised framework with active contrastive learning and human-in-the-loop feedback.

HERE is an image retrieval engine for H&E tumor pathology images that uses AI encoding and indexing to efficiently find similar images across large datasets.

This paper develops a deep learning model to predict molecular subtypes and outcomes in muscle-invasive bladder cancer using histopathological images instead of RNA sequencing.

The paper develops SINDI, a framework combining deep learning and explainable AI to accurately diagnose infectious skin diseases from histopathology images, achieving high accuracy even on challenging grain-free images.

The paper introduces an AI agent that localizes bulk RNA-seq pathway activity in histopathology images using pathologist feedback and large language models.

NLTD 2.0 improves color deconvolution in histopathology by handling more than three stains without requiring optimal color vectors, enhancing reliability and accessibility through an ImageJ plugin.

PathQC predicts RNA Integrity Number and autolysis from H&E images using deep learning, achieving moderate to high correlations across tissues.

HRDPath predicts HRD status from HGSOC histopathology images using a multi-model deep learning architecture, achieving higher AUC and specificity than prior methods.

This study benchmarks five digital pathology foundation models across various tasks, showing that model size does not always correlate with performance, and recommends selecting models based on task-specific needs.

This paper introduces EXPAND, an interpretable AI model that predicts breast cancer subtypes and survival using nuclear features, achieving performance comparable to black-box models.

ESPWA is a deep learning tool developed to improve endocrine therapy precision in low-resource settings by accurately predicting ER status from H&E slides in Haitian patients.

This paper introduces a segmentation-free method for analyzing multispectral whole slide images to study tissue repair and regeneration, revealing tissue compartments and dynamics without prior labels.

The paper develops Mosaic, a machine learning system that classifies tumors into detailed subtypes and identifies molecular phenotypes using histopathologic and genomic data from a large patient cohort.

CARP3D uses deep learning to prioritize high-risk 2D sections in 3D pathology data, improving disease detection while reducing pathologist workload.

SpotWhisperer links histopathology images to spatial gene expression and text annotations using natural language AI, outperforming existing models on a new benchmark dataset.

MiroSCOPE is an AI platform for annotating functional tissue units in digital pathology, enabling efficient annotation of over 71,900 units across prostate cancer samples.

This study improves cell type classification in lung cancer tissue sections by incorporating microenvironment information through fisheye transformations, achieving human-level accuracy.

XpressO-Melanoma predicts BRAF V600E mutation status from whole-slide images using deep learning, achieving 0.79 AUC and correctly identifying 7/8 mutant cases.

AdSI-MIMO improves stain imputation for multiplex immunofluorescence images by generating multiple biomarker outputs from varied input combinations, achieving better performance for CD8 and PD-L1 imputation.

HistoTME-v2 predicts cell type-specific transcriptomic signatures and spatial distributions from H&E images, outperforming baselines across 25 tumor types and multiple validation datasets.

This study shows that a pretrained tissue patch encoder can effectively encode cellular graphs, capturing cell-level features and spatial relationships in tissue patches.

This paper develops a deep learning model for cervical cancer classification using Pap smear images, achieving high accuracy and outperforming previous methods.

MARQO pipeline automates single-cell analysis of multiplex immunostained tissue slides, showing improved accuracy over manual methods in cancer studies.

This study develops a deep learning model to predict survival across multiple squamous tumor types using H&E stained slides, showing its predictive power beyond head and neck cancer.

This paper extends pre-training of histopathology foundation models to specialize them for tumor tissue, revealing two coexisting breast cancer archetypes defined by RNA splicing and TGF-β dysregulation.

Delta-Marches uses generative AI to simulate morphological changes between disease classes, identifying subcellular features that drive disease progression in histopathological images.

This paper introduces a deep learning model to predict collagen from H&E-stained images, enabling collagen quantification across cancer types and linking it to clinical outcomes.

The paper develops a self-supervised deep learning framework to analyze multiplexed cancer tissue microscopy images, stratifying patients into prognostically distinct groups with different survival outcomes.