Deep Gaussian Process with Uncertainty Estimation for Microsatellite Instability and Immunotherapy Response Prediction Based on Histology DOI Open Access
Sunho Park,

Mark Pettigrew,

Yoon Jin

et al.

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

Published: Nov. 3, 2024

Abstract Determining tumor microsatellite status has significant clinical value because tumors that are instability-high (MSI-H) or mismatch repair deficient (dMMR) respond well to immune check-point inhibitors (ICIs) and oftentimes not chemotherapeutics. We propose MSI-SEER, a deep Gaussian process-based Bayesian model analyzes H&E whole-slide images in weakly-supervised-learning predict gastric colorectal cancers. performed extensive validation using multiple large datasets comprised of patients from diverse racial backgrounds. MSI-SEER achieved state-of-the-art performance with MSI prediction, which was by integrating uncertainty prediction. high accuracy for predicting ICI responsiveness combining stroma-to-tumor ratio. Finally, MSI-SEER’s tile-level predictions revealed novel insights into the role spatial distribution MSI-H regions microenvironment response.

Language: Английский

MultiResFF‐Net: Multilevel Residual Block‐Based Lightweight Feature Fused Network With Attention for Gastrointestinal Disease Diagnosis DOI Creative Commons
Sohaib Asif,

Yajun Ying,

Tingting Qian

et al.

International Journal of Intelligent Systems, Journal Year: 2025, Volume and Issue: 2025(1)

Published: Jan. 1, 2025

Accurate detection of gastrointestinal (GI) diseases is crucial due to their high prevalence. Screening often inefficient with existing methods, and the complexity medical images challenges single‐model approaches. Leveraging diverse model features can improve accuracy simplify detection. In this study, we introduce a novel deep learning tailored for diagnosis GI through analysis endoscopy images. This innovative model, named MultiResFF‐Net, employs multilevel residual block‐based feature fusion network. The key strategy involves integration from truncated DenseNet121 MobileNet architectures. not only optimizes model’s diagnostic performance but also strategically minimizes computational demands, making MultiResFF‐Net valuable tool efficient accurate disease in A pivotal component enhancing introduction Modified MultiRes‐Block (MMRes‐Block) Convolutional Block Attention Module (CBAM). MMRes‐Block, customized component, optimally handles fused at endpoint both models, fostering richer sets without escalating parameters. Simultaneously, CBAM ensures dynamic recalibration maps, emphasizing relevant channels spatial locations. dual incorporation significantly reduces overfitting, augments precision, refines extraction process. Extensive evaluations on three datasets—endoscopic images, GastroVision data, histopathological images—demonstrate exceptional 99.37%, 97.47%, 99.80%, respectively. Notably, achieves superior efficiency, requiring 2.22 MFLOPS 0.47 million parameters, outperforming state‐of‐the‐art models cost‐effectiveness. These results establish as robust practical

Language: Английский

Citations

0
Predicting treatment response in multicenter non-small cell lung cancer patients based on federated learning DOI Creative Commons
Yuan Liu,

Jinzao Huang,

Jyh‐Cheng Chen

et al.

BMC Cancer, Journal Year: 2024, Volume and Issue: 24(1)

Published: June 5, 2024

Abstract Background Multicenter non-small cell lung cancer (NSCLC) patient data is information-rich. However, its direct integration becomes exceptionally challenging due to constraints involving different healthcare organizations and regulations. Traditional centralized machine learning methods require centralizing these sensitive medical for training, posing risks of privacy leakage security issues. In this context, federated (FL) has attracted much attention as a distributed framework. It effectively addresses contradiction by preserving locally, conducting local model aggregating parameters. This approach enables the utilization multicenter with maximum benefit while ensuring safeguards. Based on pre-radiotherapy planning target volume images NSCLC patients, treatment response prediction designed FL predicting probability remission patients. ensures privacy, high accuracy computing efficiency, offering valuable insights clinical decision-making. Methods We retrospectively collected CT from 245 patients undergoing chemotherapy radiotherapy (CRT) in four Chinese hospitals. simulation environment, we compared performance deep (DL) that using two sites. Additionally, unavailability one hospital, established real-world three Assessments were conducted measures such accuracy, receiver operating characteristic curve, confusion matrices. Results The model’s obtained outperforms traditional methods. comparative experiment, DL achieves an AUC 0.718/0.695, demonstrates 0.725/0.689, achieving 0.698/0.672. Conclusions demonstrate predictive model, developed combining convolutional neural networks (CNNs) multiple centers, comparable through training. can efficiently predict CRT privacy.

Language: Английский

Citations

2
Brain Storm Optimization based Swarm Learning for Diabetic Retinopathy Image Classification DOI
Liang Qu, Cunze Wang, Yuhui Shi

et al.

2022 IEEE Congress on Evolutionary Computation (CEC), Journal Year: 2024, Volume and Issue: unknown, P. 1 - 7

Published: June 30, 2024

Language: Английский

Citations

1
Improving the Annotation Process in Computational Pathology: A Pilot Study with Manual and Semi-automated Approaches on Consumer and Medical Grade Devices DOI Creative Commons
Giorgio Cazzaniga,

Fabio Del Carro,

Albino Eccher

et al.

Deleted Journal, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 4, 2024

Abstract The development of reliable artificial intelligence (AI) algorithms in pathology often depends on ground truth provided by annotation whole slide images (WSI), a time-consuming and operator-dependent process. A comparative analysis different approaches is performed to streamline this Two pathologists annotated renal tissue using semi-automated (Segment Anything Model, SAM)) manual devices (touchpad vs mouse). comparison was conducted terms working time, reproducibility (overlap fraction), precision (0 10 accuracy rated two expert nephropathologists) among methods operators. impact displays mouse performance evaluated. Annotations focused three compartments: tubules (57 annotations), glomeruli (53 arteries (58 annotations). semi-automatic approach the fastest had least inter-observer variability, averaging 13.6 ± 0.2 min with difference ( Δ ) 2%, followed (29.9 10.2, = 24%), touchpad (47.5 19.6 min, 45%). highest achieved SAM values 1 0.99 compared 0.97 for 0.94 0.93 touchpad), though lower value 0.89 both touchpad). No differences were observed between operators p 0.59). Using non-medical monitors increased times 6.1%. future employment AI-assisted can significantly speed up process, improving AI tool development.

Language: Английский

Citations

1
Accuracy of machine learning in diagnosing microsatellite instability in gastric cancer: A systematic review and meta-analysis DOI Creative Commons
Yiming Ying, Robert J. Ju,

Jieyi Wang

et al.

International Journal of Medical Informatics, Journal Year: 2024, Volume and Issue: 193, P. 105685 - 105685

Published: Nov. 2, 2024

Significant challenges persist in the early identification of microsatellite instability (MSI) within current clinical practice. In recent years, with growing utilization machine learning (ML) diagnosis and management gastric cancer (GC), numerous researchers have explored effectiveness ML methodologies detecting MSI. Nevertheless, predictive value these approaches still lacks comprehensive evidence. Accordingly, this study was carried out to consolidate accuracy prompt detection MSI GC.

Language: Английский

Citations

1
Prediction of Microsatellite Instability From Gastric Histological Images Based on Residual Attention Networks With Non-Local Modules DOI Creative Commons
Sung‐Nien Yu, Shih‐Chiang Huang, Weichen Wang

et al.

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 60374 - 60385

Published: Jan. 1, 2023

Gastric cancer can be classified into different subtypes according to their genetic expression. Microsatellite instability (MSI) is one of these and an important clinical marker for prognosis consideration immunotherapy. Since testing relatively expensive laborious, this study tackles the challenge using deep neural networks (DNNs) identify MSI based on analyzing histomorphologic features gastric whole-slide images (WSIs). A two-stage patch-wise framework was proposed, which first differentiates tumor regions from normal, then predicts status tumorous patches. The proposed learning architecture enhances residual attention network with non-local modules visual context fusion modules, thereby allowing both local fine-grained details coarse long-range dependencies captured. Image post-processing procedures were also better align region segmentation pathologist annotations. model applied a three-way classification task, namely normal tissue, microsatellite stable (MSS), MSI, private dataset gathered by Chang Gung Memorial Hospital achieved 91.95% slide-wise accuracy. We studied feasibility transfer fine tuning TCGA-STAD public dataset, where we attained high accuracy 96.53% AUC 0.99, outperforming previous literature.

Language: Английский

Citations

3
Artificial Intelligence in Respiratory Medicine DOI Creative Commons

K Kalaiyarasan,

Rajeswari Sridhar

Journal of Association of Pulmonologist of Tamil Nadu, Journal Year: 2023, Volume and Issue: 6(2), P. 53 - 68

Published: May 1, 2023

The integration of artificial intelligence (AI) and the medical field has opened a wide range possibilities. Currently, role AI in is limited to image analysis (radiological histopathology images), identifying alerting about specific health conditions, supporting clinical decisions. future lung cancer screening, diagnosis, management expected undergo significant transformation with use radiomics, radiogenomics, virtual biopsy. can also help physicians diagnose treat variety respiratory illnesses, including interstitial diseases, asthma, chronic obstructive pulmonary disease, pleural diseases such as effusion pneumothorax, pneumonia, artery hypertension, tuberculosis. automated reporting function tests, polysomnography, recorded breath sounds. Through robotic technology, set create new milestones realm interventional pulmonology. A well-trained may offer insights into genetic molecular mechanisms pathogenesis various assist outlining best course action horizontal patients' digital records, radiographic images, pathology biochemical lab reports. As any doctors researchers should be aware advantages limitations AI, they it responsibly advance knowledge provide better care patients.

Language: Английский

Citations

3
Artificial intelligence and digital pathology as drivers of precision oncology DOI
Yuri Tolkach,

Sebastian Klein,

Tsvetan Tsvetkov

et al.

Deleted Journal, Journal Year: 2023, Volume and Issue: 29(10), P. 839 - 850

Published: June 6, 2023

Die Digitalisierung bietet viele Chancen zur Verbesserung von Diagnostik und Therapien bei Krebserkrankungen, insbesondere auch im Bereich der Pathologie. Neben molekularen Analyse bösartigen Tumoren Proteinanalytik (Immunhistochemie) ist die Pathologie ein weiterer evolutionärer Schritt, dieses Fachgebiet tiefgreifend modernisieren transformieren wird. vorliegende Arbeit basiert auf einer selektiven Literaturrecherche in Datenbank PubMed zum Thema "digitale Pathologie" "KI-Algorithmen". Das Spektrum digitalen Transformation reicht Strukturierung diagnostischer Befunde schnelleren, präziseren reproduzierbareren für onkologische Patientinnen Patienten über Einführung Telepathologie einen schnelleren Zugang zu Referenzpathologien bis hin Algorithmen, künstlicher Intelligenz (KI) beruhen automatisierte Analysen virtualisierter pathologischer Gewebsschnitte ermöglichen. Letztere sind Gegenstand aktiver Forschung gliedern sich 2 Hauptkategorien: i.) diagnostische KI typische Aufgaben Pathologie, beispielsweise Quantifizierung prädiktiver Marker immunhistochemischer Färbungen oder Tumordetektionen, Graduierungen Subtypisierungen anhand Hämatoxylin-Eosin-Routinefärbungen, sowie ii.) fortgeschrittene Anwendungen, welche Detektionen molekulargenetischen Alterationen therapierelevante Bildbiomarker, weitesten Sinne, beinhalten. In dieser werden Aspekte pathologische Institute reflektiert, Hinblick zukünftige Entwicklung Präzisionsonkologie, aber deren Status quo.

Citations

1
Artificial intelligence transforms the future of oncology care DOI
Archana Behera,

Mukesh Kumar Dharmalingam Jothinathan

Journal of Stomatology Oral and Maxillofacial Surgery, Journal Year: 2024, Volume and Issue: 125(4), P. 101915 - 101915

Published: May 16, 2024

Language: Английский

Citations

0
Swarm mutual learning DOI Creative Commons

Kang Hai-yan,

Wang Jiakang

Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(6), P. 8063 - 8077

Published: Aug. 14, 2024

With the rapid growth of big data, extracting meaningful knowledge from data is crucial for machine learning. The existing Swarm Learning collaboration models face challenges such as security, model high communication overhead, and performance optimization. To address this, we propose Mutual (SML). Firstly, introduce an Adaptive Distillation Algorithm that dynamically controls learning intensity based on distillation weights strength, enhancing efficiency extraction transfer during mutual distillation. Secondly, design a Global Parameter Aggregation homomorphic encryption, coupled with Dynamic Gradient Decomposition using singular value decomposition. This allows to aggregate parameters in ciphertext, significantly reducing overhead uploads downloads. Finally, validate proposed methods real datasets, demonstrating their effectiveness updates. On MNIST dataset CIFAR-10 dataset, local accuracies reached 95.02% 55.26%, respectively, surpassing those comparative models. Furthermore, while ensuring security aggregation process, reduced uploading downloading.

Language: Английский

Citations

0