Improving unsupervised stain-to-stain translation using self-supervision and meta-learning

Bouteldja, Nassim; Klinkhammer, Barbara M.; Schlaich, Tarek; Merhof, Dorit

doi:10.1016/j.jpi.2022.100107

Cited by 17 publications

(22 citation statements)

References 24 publications

(52 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We then resampled the detected tissue into 0.337 μm pixel spacing and tessellated it into images of size 640 x 640 pixels to comply with the input requirements of the pretrained segmentation model. The CycleGANs were then trained on the extracted tiles using the same architecture and training routine from Bouteldja et al, 10 which showed promising performance for domain adaptation across stains. In short, U-Net-like generators with a depth of seven and PatchGAN 32 discriminators with a depth of 4 were trained for 300 000 iterations using RAdam on random minibatches of size 3.…”

Section: Methodsmentioning

confidence: 99%

“…reducing color variations within a specific stain. Second, stain translation, 10 , 11 , 12 , 13 , 14 i.e. converting between different stains.…”

Section: Introductionmentioning

confidence: 99%

“…Several works have also included a feature-preserving loss on feature maps or outputs by a prior CNN to improve the image translation. 6 , 7 , 10 De Bel et al 9 trained a CNN for the segmentation of glomerular tufts on one data center (Amsterdam) and applied it to another center (Radboud) using extensive color augmentation during training and CycleGANs for stain normalization. They found improvements in segmentation performance when additionally employing CycleGAN-based stain normalization on top of extensive color augmentations.…”

Section: Introductionmentioning

confidence: 99%

“…the glomerular tuft, although CycleGAN-based image translation can show highly varying effects on the segmentation of different structures. 10 …”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Tackling stain variability using CycleGAN-based stain augmentation

Bouteldja

Hölscher

Bülow

et al. 2022

Journal of Pathology Informatics

Self Cite

View full text Add to dashboard Cite

Section: Methodsmentioning

confidence: 99%

“…reducing color variations within a specific stain. Second, stain translation, 10 , 11 , 12 , 13 , 14 i.e. converting between different stains.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…the glomerular tuft, although CycleGAN-based image translation can show highly varying effects on the segmentation of different structures. 10 …”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Tackling stain variability using CycleGAN-based stain augmentation

Bouteldja

Hölscher

Bülow

et al. 2022

Journal of Pathology Informatics

Self Cite

View full text Add to dashboard Cite

“…Immunhistochemie und Immunfluoreszenz ineinander konvertieren, die Anzahl von Bildern von Fällen seltener Erkrankungen steigern oder andere Trainingsdaten für KNN erstellen. Die Translation beispielsweise einer immunhistochemischen Färbung in eine PAS-Färbung [ 5 ] ist hilfreich, wenn ein Segmentierungs-KNN nur auf PAS-Schnitten trainiert wurde. Durch die Translation in eine PAS-Färbung ist die immunhistochemische Färbung durch das KNN segmentierbar mit dem Potenzial, eine kompartimentspezifische Analyse der Immunhistochemie automatisiert durchzuführen.…”

Section: Generierung Synthetischer Datenunclassified

Automatische Bildanalyse und künstliche Intelligenz in der Nephropathologie

Bülow

Hölscher

2022

Nephrologie

Self Cite

View full text Add to dashboard Cite

Hintergrund Die digitale Transformation der Pathologie durch die weite Verbreitung von sog. Whole-slide-Scannern bietet zahlreiche Möglichkeiten für die Nephropathologie, insbesondere hinsichtlich der Implementierung von Computerassistenz. Aktuell werden intensiv die Möglichkeiten von Systemen, die auf der Nutzung von Deep Learning, einer speziellen Technik der Informationsverarbeitung, beruhen, erforscht. Fragestellung Ziel ist die Ermittlung des gegenwärtigen Standes der Forschung hinsichtlich Anwendungen von Deep-Learning-Methoden zur Bildanalyse in der Nephropathologie. Material und Methoden Literaturrecherche im Web of Science (WOS) und in Pubmed. Für Abbildung 1 wurde die folgende Suchabfrage im WOS genutzt: ALL=(Digital Pathology AND AI OR Deep Learning OR Machine Learning). Ergebnisse Es liegen zahlreiche Anwendungen von Deep-Learning-basierten Methoden zur Unterstützung in der Nephropathologie vor. Diese fokussieren im Wesentlichen auf die Segmentierung und Quantifizierung der Nierenhistologie, wobei zunehmend auch die diagnostische Klassifikation und die Generierung synthetischer Daten erforscht werden. Die Translation dieser Systeme in den diagnostischen Alltag ist gegenwärtig noch nicht erfolgt. Es fehlt dazu beispielsweise an prospektiver Evidenz, die den Nutzen dieser Methoden in der klinischen Versorgung darlegt. Diskussion Die Implementierung der digitalen Nephropathologie mit Assistenz durch auf Deep Learning basierenden Methoden hat ein großes Potenzial. Die translationale Lücke sollte in Zukunft durch multizentrische, prospektive, interdisziplinär durchgeführte Studien gefüllt werden.

show abstract

Time for a full digital approach in nephropathology: a systematic review of current artificial intelligence applications and future directions

Cazzaniga,

Rossi,

Eccher

et al. 2023

J Nephrol

View full text Add to dashboard Cite

Introduction Artificial intelligence (AI) integration in nephropathology has been growing rapidly in recent years, facing several challenges including the wide range of histological techniques used, the low occurrence of certain diseases, and the need for data sharing. This narrative review retraces the history of AI in nephropathology and provides insights into potential future developments. Methods Electronic searches in PubMed-MEDLINE and Embase were made to extract pertinent articles from the literature. Works about automated image analysis or the application of an AI algorithm on non-neoplastic kidney histological samples were included and analyzed to extract information such as publication year, AI task, and learning type. Prepublication servers and reviews were not included. Results Seventy-six (76) original research articles were selected. Most of the studies were conducted in the United States in the last 7 years. To date, research has been mainly conducted on relatively easy tasks, like single-stain glomerular segmentation. However, there is a trend towards developing more complex tasks such as glomerular multi-stain classification. Conclusion Deep learning has been used to identify patterns in complex histopathology data and looks promising for the comprehensive assessment of renal biopsy, through the use of multiple stains and virtual staining techniques. Hybrid and collaborative learning approaches have also been explored to utilize large amounts of unlabeled data. A diverse team of experts, including nephropathologists, computer scientists, and clinicians, is crucial for the development of AI systems for nephropathology. Collaborative efforts among multidisciplinary experts result in clinically relevant and effective AI tools. Graphical abstract

show abstract

Improving unsupervised stain-to-stain translation using self-supervision and meta-learning

Cited by 17 publications

References 24 publications

Tackling stain variability using CycleGAN-based stain augmentation

Tackling stain variability using CycleGAN-based stain augmentation

Automatische Bildanalyse und künstliche Intelligenz in der Nephropathologie

Time for a full digital approach in nephropathology: a systematic review of current artificial intelligence applications and future directions

Contact Info

Product

Resources

About