Automated curation of large‐scale cancer histopathology image datasets using deep learning

Histopathology, Journal Year: 2024, Volume and Issue: 84(7), P. 1139 - 1153

Published: Feb. 26, 2024

Artificial intelligence (AI) has numerous applications in pathology, supporting diagnosis and prognostication cancer. However, most AI models are trained on highly selected data, typically one tissue slide per patient. In reality, especially for large surgical resection specimens, dozens of slides can be available each Manually sorting labelling whole-slide images (WSIs) is a very time-consuming process, hindering the direct application collected samples from cohorts. this study we addressed issue by developing deep-learning (DL)-based method automatic curation pathology datasets with several

Language: Английский

---

A combination between transfer learning models and UNet++ for COVID-19 diagnosis

Multimedia Tools and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 9, 2024

Language: Английский

---

Digital and Computational Pathology Are Pathologists’ Physician Extenders

Archives of Pathology & Laboratory Medicine, Journal Year: 2024, Volume and Issue: 148(8), P. 866 - 870

Published: March 27, 2024

Language: Английский

---

Assessment of PD-L1 expression and tumour infiltrating lymphocytes in early-stage non-small cell lung carcinoma with artificial intelligence algorithms

Journal of Clinical Pathology, Journal Year: 2024, Volume and Issue: unknown, P. jcp - 209766

Published: Oct. 17, 2024

To study programmed death ligand 1 (PD-L1) expression and tumour infiltrating lymphocytes (TILs) in patients with early-stage non-small cell lung carcinoma (NSCLC) artificial intelligence (AI) algorithms.

Language: Английский

---

Prediction models for hormone receptor status in female breast cancer do not extend to males: further evidence of sex-based disparity in breast cancer

npj Breast Cancer, Journal Year: 2023, Volume and Issue: 9(1)

Published: Nov. 8, 2023

Breast cancer prognosis and management for both men women are reliant upon estrogen receptor alpha (ERα) progesterone (PR) expression to inform therapy. Previous studies have shown that there sex-specific binding characteristics of ERα PR in breast and, counterintuitively, is more common male than female cancer. We hypothesized these differences could morphological manifestations undetectable human observers but be elucidated computationally. To investigate this, we trained attention-based multiple instance learning prediction models using H&E-stained images from the Cancer Genome Atlas (TCGA) (n = 1085) deployed them on external 192) 245). Both targets were predicted internal (AUROC prediction: 0.86 ± 0.02, p < 0.001; AUROC 0.76 0.03, 0.001) cohorts 0.78 0.80 0.04, not cohort 0.66 0.14, 0.43; 0.63 0.05). This suggests subtle invisible visual inspection may exist between sexes, supporting previous immunohistochemical, genomic, transcriptomic analyses.

Language: Английский

---

Patologia digital e computacional: um panorama atual e perspectivas futuras

Contribuciones a las Ciencias Sociales, Journal Year: 2024, Volume and Issue: 17(4), P. e6027 - e6027

Published: April 12, 2024

Determinar, a partir da literatura vigente, as evidências científicas sobre o uso patologia digital e computacional na prática clínica, considerando benefícios, limitações perspectivas futuras. Scoping Review fundamentada nas diretrizes do Joanna Briggs Institute. Pesquisa conduzida, entre maio julho de 2023, no banco dados Medline, listas referências cinzenta adicionais. Dois revisores independentes selecionaram títulos resumos acordo com os critérios inclusão. Foram considerados estudos que reportassem aspectos relevantes computacional. Evidenciou-se avanços obtenção imagem lâmina inteira, geradas por microscópios robóticos. As imagens obtidas apresentam alta precisão diagnóstica, em comparação aos exames convencionais, além possibilidade armazenamento compartilhamento destes dados. Consequentemente, destaca-se evolução computacional, implementa inteligência artificial para diagnósticos assistidos computador. A clínica está passando um grande salto evolutivo mundial, qual representa uma fase transição modelo incorpora ferramentas aprendizado máquina. Os tecnologia digitalização aprimoraram pesquisa baseada tecidos meio microscopia análise imagens, ampliando relevância sensibilidade dos achados diagnósticos.

---

Adaptive Compression Framework for Giga-pixel Whole Slide Images

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: July 30, 2024

Digital pathology images require significant storage space, leading to high costs. To address this, compressing for and restoration has been proposed; however, this involves a trade-off between compression ratio information loss. Traditional techniques often apply uniform ratio, ignoring the variable informational content across different slide regions (information disequilibrium). We proposed an Adaptive framework giga-pixel whole Slide images, namely AdaSlide, overcomes limitation by integrating decision agent (CDA) foundational image enhancer (FIE), enabling adaptive decisions aware of disequilibrium. The CDA uses reinforcement learning assess each patch's necessity degree compression, ensuring minimal loss maintaining diagnostic integrity. FIE, trained on diverse cancer types magnifications, guarantees high-quality post-compression restoration. FIE's performance was evaluated using visual Turing test, where experts could barely distinguish real compressed-restored (55% accuracy, coincidence level: 50%). In six downstream tasks (including patch-level classification, segmentation, slide-level classification), AdaSlide maintained prediction original in five out tasks. contrast, traditional methods with only two tasks, raising concerns about Additionally, store data less than 10% WSI size. This indicates that, unlike methods, can efficiently compress while preserving clinically information. Furthermore, provides flexibility its study objective-oriented reward function, tendency, FIE backbone architectures. approach ensures efficient retrieval, potentially transforming management digital systems aligning strategies clinical relevance, thereby facilitating both cost reduction improved processes.

Language: Английский

---

The Robustness of Deep Learning Models to Adversarial Attacks in Lung X-ray Classification

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 26, 2024

With the rapid advancement of artificial intelligence (AI) and deep learning, AI-driven models are increasingly being used in medical field for disease classification diagnosis. However, robustness these against adversarial attacks is a critical concern, as such can significantly distort diagnostic outcomes, leading to potential clinical errors. This study investigates various convolutional neural network (CNN) models, including MobileNet, Resnet-152, Vision Transformers (ViT), lung radiograph tasks under conditions. We utilized "ChestX-ray8" dataset train evaluate applying range attack methods, FGSM AutoAttack, assess models' resilience. Our findings indicate that while all experienced decrease accuracy after attacks, MobileNet consistently demonstrated superior compared other CNN-based models. also explored impact inverse training enhance model stability. Results seem prove sparser nature parameters, reason its robustness, will give insight into enhancement security dependability within AI applications. research underscores need continued refinement ensure their safe deployment settings.

Language: Английский

---

Artificial Intelligence Nomenclature Identified From Delphi Study on Key Issues Related to Trust and Barriers to Adoption for Autonomous Systems

Published: Sept. 13, 2023

<p>The rapid integration of artificial intelligence across traditional research domains has generated an amalgamation nomenclature. As cross-discipline teams work together on complex machine learning challenges, finding a consensus basic definitions in the literature is more fundamental problem. step Delphi process to define issues with trust and barriers adoption autonomous systems, our study first collected ranked top concerns from panel international experts fields engineering, computer science, medicine, aerospace, defence, experience working intelligence. This document presents summary for nomenclature derived expert feedback.</p>

Language: Английский

---

Results of the European Society of Toxicologic Pathology Survey on the Use of Artificial Intelligence in Toxicologic Pathology

Toxicologic Pathology, Journal Year: 2023, Volume and Issue: 51(4), P. 216 - 224

Published: June 1, 2023

The European Society of Toxicologic Pathology (ESTP) initiated a survey through its 2.0 workstream in partnership with sister professional societies Europe and North America to generate snapshot artificial intelligence (AI) usage the field toxicologic pathology. In addition demographic information, some general questions explored AI relative (1) current status adoption across organizations; (2) technical methodological aspects; (3) perceived business value finally; (4) roadblocks perspectives. has become increasingly established pathology most pathologists being supportive development despite areas uncertainty. A salient feature consisted variability awareness among responders, as spectrum extended from having developed familiarity skills AI, colleagues who had no interest tool Despite enthusiasm for these techniques, overall understanding trust algorithms well their added were generally low, suggesting room need increased education. This will serve basis evaluate evolution penetration acceptance this domain.

Language: Английский

---