Artificial intelligence in drug development

Nature Medicine, Journal Year: 2025, Volume and Issue: 31(1), P. 45 - 59

Published: Jan. 1, 2025

Language: Английский

---

The Histomorphology to Molecular Transition: Exploring the Genomic Landscape of Poorly Differentiated Epithelial Endometrial Cancers

Cells, Journal Year: 2025, Volume and Issue: 14(5), P. 382 - 382

Published: March 5, 2025

The peremptory need to circumvent challenges associated with poorly differentiated epithelial endometrial cancers (PDEECs), also known as Type II (ECs), has prompted therapeutic interrogation of the prototypically intractable and most prevalent gynecological malignancy. PDEECs account for cancer-related mortalities due their aggressive nature, late-stage detection, poor response standard therapies. are characterized by heterogeneous histopathological features distinct molecular profiles, they pose significant clinical propensity rapid progression. Regardless complexities around PDEECs, still being administered inefficiently in same manner clinically indolent readily curable type-I ECs. Currently, there no targeted therapies treatment PDEECs. realization new options transformed our understanding enabling more precise classification based on genomic profiling. transition from a provided critical insights into underlying genetic epigenetic alterations these malignancies. This review explores landscape focus identifying key subtypes mutations that variants. Here, we discuss how correlates outcomes can refine diagnostic accuracy, predict patient prognosis, inform strategies. Deciphering underpinnings led advances precision oncology protracted remissions patients untamable

Language: Английский

---

Morphological Prototyping for Unsupervised Slide Representation Learning in Computational Pathology

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2024, Volume and Issue: 3, P. 11566 - 11578

Published: June 16, 2024

Language: Английский

---

Self-Supervised Learning Reveals Clinically Relevant Histomorphological Patterns for Therapeutic Strategies in Colon Cancer

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: March 4, 2024

Abstract Self-supervised learning (SSL) automates the extraction and interpretation of histopathology features on unannotated hematoxylin-and-eosin-stained whole-slide images (WSIs). We trained an SSL Barlow Twins-encoder 435 TCGA colon adenocarcinoma WSIs to extract from small image patches. Leiden community detection then grouped tiles into histomorphological phenotype clusters (HPCs). HPC reproducibility predictive ability for overall survival was confirmed in independent clinical trial cohort (N=1213 WSIs). This unbiased atlas resulted 47 HPCs displaying unique sharing clinically significant traits, highlighting tissue type, quantity, architecture, especially context tumor stroma. Through in-depth analysis these HPCs, including immune landscape gene set enrichment analysis, association outcomes, we shed light factors influencing responses treatments like standard adjuvant chemotherapy experimental therapies. Further exploration may unveil new insights aid decision-making personalized cancer patients.

Language: Английский

---

Self-Supervised Learning Can Distinguish Myelodysplastic Neoplasms from Clinical Mimics Using Bone Marrow Biopsies

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 21, 2025

Abstract The diagnosis of myelodysplastic neoplasms (MDS) requires examination the bone marrow for morphologic evidence dysplasia. We sought to determine if a self-supervised learning (SSL) AI image analysis approach may be utilized reliably distinguish MDS from its clinically relevant mimics using biopsies (BMBx). Whole slide images (WSIs) H&E- and reticulin-stained BMBx sections 243 unique patients (89 MDS, 55 non-MDS cytopenic controls [NMCC], 99 negative control [NC] cases) were segmented into tiles analyzed. These then processed Barlow Twins SSL model generate histomorphologic phenotype clusters (HPCs). Review HPCs revealed enriched in captured known histopathologic features including hypercellularity, dysplastic clustered megakaryocytes, increased immature hematopoietic cells, vascularity, fibrosis, cell streaming patterns. Assessment 95 second institution showed consistent HPC enrichment patterns, validating robustness model. trained ensemble slides distinguished NCs with an AUC 0.89, age-matched, NMCCs 0.84. findings demonstrate potential approaches capture diagnostically patterns improve reproducibility diagnosis.

Language: Английский

---

Self-supervised learning reveals clinically relevant histomorphological patterns for therapeutic strategies in colon cancer

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: March 8, 2025

Abstract Self-supervised learning (SSL) automates the extraction and interpretation of histopathology features on unannotated hematoxylin-eosin-stained whole slide images (WSIs). We train an SSL Barlow Twins encoder 435 colon adenocarcinoma WSIs from The Cancer Genome Atlas to extract small image patches (tiles). Leiden community detection groups tiles into histomorphological phenotype clusters (HPCs). HPC reproducibility predictive ability for overall survival are confirmed in independent clinical trial ( N = 1213 WSIs). This unbiased atlas results 47 HPCs displaying unique shared clinically significant traits, highlighting tissue type, quantity, architecture, especially context tumor stroma. Through in-depth analyses these HPCs, including immune landscape gene set enrichment analyses, associations outcomes, we shine light factors influencing responses treatments standard adjuvant chemotherapy experimental therapies. Further exploration may unveil additional insights aid decision-making personalized cancer patients.

Language: Английский

---

HistoMoCo: Momentum Contrastive Learning Pre-Training on Unlabeled Histopathological Images for Oral Squamous Cell Carcinoma Detection

Electronics, Journal Year: 2025, Volume and Issue: 14(7), P. 1252 - 1252

Published: March 22, 2025

The early detection and intervention of oral squamous cell carcinoma (OSCC) using histopathological images are crucial for improving patient outcomes. current literature identifying OSCC predominantly relies on models pre-trained ImageNet to minimize the need manual data annotations in model fine-tuning. However, a significant divergence exists between visual domains natural images, potentially limiting representation transferability these models. Inspired by recent self-supervised research, this work, we propose HistoMoCo, an adaptation Momentum Contrastive Learning (MoCo), designed generate with enhanced image representations initializations images. Specifically, HistoMoCo aggregates 102,228 leverages structure features unique histological data, allowing more robust feature extraction subsequent downstream We perform tasks evaluate two real-world datasets, including NDB-UFES Oral Histopathology datasets. Experimental results demonstrate that consistently outperforms traditional ImageNet-based pre-training, yielding stable accurate performance detection, achieving AUROC up 99.4% dataset 94.8% dataset. Furthermore, dataset, pre-training solution achieves 89.32% 40% training whereas reaches 89.58% only 10% data. addresses issue domain state-of-the-art More importantly, significantly reduces reliance release our code parameters further research histopathology or tasks.

Language: Английский

---

Integrated multicenter deep learning system for prognostic prediction in bladder cancer

npj Precision Oncology, Journal Year: 2024, Volume and Issue: 8(1)

Published: Oct. 16, 2024

Precise survival risk stratification is crucial for personalized therapy in bladder cancer (BCa). This study developed and validated an end-to-end deep learning system using histological slides to predict overall (OS) BCa patients. We employed the BlaPaSeg tile classifier generate tissue probability heatmaps segmentation maps, trained two prognostic networks, MacroVisionNet UniVisionNet, explored six potential biomarkers. Across all cohorts, AUC ranged from 0.9906 0.9945, while C-index varied 0.655 0.834 0.661 0.853 UniVisionNet. After covariate adjustment, hazard ratio (HR) values high-risk groups were 1.97 5.06 2.13 4.01 The Coloc (Tumor Co-localization score) IMTS (Integrated Muscle Tumor Score) illustrated a higher death with HR 1.41 10.16. improves prediction supports refined patient management.

Language: Английский

---

Deep learning model for automated diagnosis of moyamoya disease based on magnetic resonance angiography

EClinicalMedicine, Journal Year: 2024, Volume and Issue: 77, P. 102888 - 102888

Published: Nov. 1, 2024

SummaryBackgroundThis study explores the potential of deep learning-based convolutional neural network (CNN) to automatically recognize MMD using MRA images from atherosclerotic disease (ASD) and normal control (NC).MethodsIn this retrospective in China, 600 participants (200 MMD, 200 ASD NC) were collected one institution as an internal dataset for training 60 another external testing set validation. All divided into (N = 450) validation sets 90), 60), 60). The input CNN models comprised preprocessed images, while output was a tripartite classification label that identified patient's diagnostic group. performances 3D evaluated comprehensive metrics such area under curve (AUC) accuracy. Gradient-weighted Class Activation Mapping (Grad-CAM) used visualize CNN's decision-making process diagnosis by highlighting key areas. Finally, compared with those two experienced radiologists.FindingsDenseNet-121 exhibited superior discrimination capabilities, achieving macro-average AUC 0.977 (95% CI, 0.928–0.995) test 0.880 0.786–0.937) sets, thus exhibiting comparable capabilities human radiologists. In binary where NC group together, separate targeted detection, DenseNet-121 achieved accuracy 0.967 0.886–0.991). Additionally, Grad-CAM results areas intense redness indicating critical model, reflected similar experts.InterpretationThis highlights efficacy model automated on easing workload radiologists promising integration clinical workflows.FundingNational Natural Science Foundation Tianjin Technology Project Beijing Foundation.

Language: Английский

---

Artificial intelligence in lung cancer: current applications, future perspectives, and challenges

Frontiers in Oncology, Journal Year: 2024, Volume and Issue: 14

Published: Dec. 23, 2024

Artificial intelligence (AI) has significantly impacted various fields, including oncology. This comprehensive review examines the current applications and future prospects of AI in lung cancer research treatment. We critically analyze latest technologies their across multiple domains, genomics, transcriptomics, proteomics, metabolomics, immunomics, microbiomics, radiomics, pathomics research. The elucidates AI’s transformative role enhancing early detection, personalizing treatment strategies, accelerating therapeutic innovations. explore impact on precision medicine cancer, encompassing diagnosis, planning, monitoring, drug discovery. potential analyzing complex datasets, genetic profiles, imaging data, clinical records, is discussed, highlighting its capacity to provide more accurate diagnoses tailored plans. Additionally, we examine predicting patient responses immunotherapy forecasting survival rates, particularly non-small cell (NSCLC). addresses technical challenges facing implementation care, data quality quantity issues, model interpretability, ethical considerations, while discussing solutions emphasizing importance rigorous validation. By providing a analysis for researchers clinicians, this underscores indispensable combating usher new era medical breakthroughs, ultimately aiming improve outcomes life.

Language: Английский

---