An Ensemble of 3D U-Net Based Models for Segmentation of Kidney and Masses in CT Scans

Golts, Alex; Khapun, Daniel; Shats, Daniel; Shoshan, Yoel; Gilboa-Solomon, Flora

doi:10.1007/978-3-030-98385-7_14

Cited by 7 publications

(6 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tumors in the MR dataset were on average smaller than those in the CT dataset since the images came from patients undergoing partial and radical nephrectomies. In addition, we did not evaluate different configurations in the nnU-Net framework nor a series of different nnU-Net models as employed in some KiTS21 submissions [16,17]. The last major limitation of this work is that this method establishes plateau performance relative to the holdout test set, which segmentation model developers must independently ensure is representative of the real-world images for the intended task.…”

Section: Discussionmentioning

confidence: 99%

“…The nnU-Net framework is recognized as the state-of-the-art framework in medical image semantic segmentation, being externally validated and winning several open-sourced medical image segmentation challenges in the 2018 Medical Decathlon Segmentation Challenge [15]. Additionally, all the top submissions in the 2021 open-access Kidney and Kidney Tumor Segmentation Challenges used variations of the nnU-Net framework [16,17].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Effect of Dataset Size and Medical Image Modality on Convolutional Neural Network Model Performance for Automated Segmentation: A CT and MR Renal Tumor Imaging Study

et al. 2023

View full text Add to dashboard Cite

The aim of this study is to investigate the use of an exponential-plateau model to determine the required training dataset size that yields the maximum medical image segmentation performance. CT and MR images of patients with renal tumors acquired between 1997 and 2017 were retrospectively collected from our nephrectomy registry. Modality-based datasets of 50, 100, 150, 200, 250, and 300 images were assembled to train models with an 80–20 training-validation split evaluated against 50 randomly held out test set images. A third experiment using the KiTS21 dataset was also used to explore the effects of different model architectures. Exponential-plateau models were used to establish the relationship of dataset size to model generalizability performance. For segmenting non-neoplastic kidney regions on CT and MR imaging, our model yielded test Dice score plateaus of $$0.93\pm 0.02$$ 0.93 ± 0.02 and $$0.92\pm 0.04$$ 0.92 ± 0.04 with the number of training-validation images needed to reach the plateaus of 54 and 122, respectively. For segmenting CT and MR tumor regions, we modeled a test Dice score plateau of $$0.85\pm 0.20$$ 0.85 ± 0.20 and $$0.76\pm 0.27$$ 0.76 ± 0.27 , with 125 and 389 training-validation images needed to reach the plateaus. For the KiTS21 dataset, the best Dice score plateaus for nn-UNet 2D and 3D architectures were $$0.67\pm 0.29$$ 0.67 ± 0.29 and $$0.84\pm 0.18$$ 0.84 ± 0.18 with number to reach performance plateau of 177 and 440. Our research validates that differing imaging modalities, target structures, and model architectures all affect the amount of training images required to reach a performance plateau. The modeling approach we developed will help future researchers determine for their experiments when additional training-validation images will likely not further improve model performance.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Effect of Dataset Size and Medical Image Modality on Convolutional Neural Network Model Performance for Automated Segmentation: A CT and MR Renal Tumor Imaging Study

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Before open sourcing it, we used fuse internally in multiple research projects (Raboh, Levanony, et al, 2022), (Rabinovici-Cohen, Tlusty, et al, 2022, (Jubran et al, 2021), (Tlusty et al, 2021), (Golts et al, 2022), (Barros et al) and experienced significant improvement in development time, reusability and collaboration. We were also able to meaningfully measure our progress and statistical significance of our results with off-the-shelf fuse.eval components that facilitate metrics' confidence interval calculation and model comparison.…”

Section: Statement Of Needmentioning

confidence: 99%

FuseMedML: a framework for accelerated discovery in machine learning based biomedicine

Golts¹,

Raboh²,

Shoshan³

et al. 2023

JOSS

Self Cite

View full text Add to dashboard Cite

Machine Learning is at the forefront of scientific progress in Healthcare and Medicine. To accelerate scientific discovery, it is important to have tools that allow progress iterations to be collaborative, reproducible, reusable and easily built upon without "reinventing the wheel" for each task. FuseMedML, or fuse, is a Python framework designed for accelerated Machine Learning (ML) based discovery in the medical domain. It is highly flexible and designed for easy collaboration, encouraging code reuse. Flexibility is enabled by a generic data object design where data is kept in a nested (hierarchical) Python dictionary (NDict), allowing to efficiently process and fuse information from multiple modalities. Functional components allow to specify input and output keys, to be read from and written to the nested dictionary. Easy code reuse is enabled through key components implemented as standalone packages under the main fuse repo using the same design principles. These include fuse.data -a flexible data processing pipeline, fuse.dl -reusable Deep Learning (DL) model architecture components and loss functions, and fuse.eval -a library for evaluating ML models.

show abstract

“…Abordagens que aplicam sucessivas redes de segmentac ¸ão semântica para a localizac ¸ão de objetos em imagens são denominadas como arquitetura em cascata ou emsemble. Os trabalhos [Zhao et al 2022, Golts et al 2022, George 2022] desenvolvem esta abordagem para a segmentac ¸ão de rins, cistos e tumores renais. O método proposto no primeiro trabalho é estruturado duas etapas, segmentac ¸ão geral e especifica, que aplicam uma versão modificada da arquitetura U-Net, denominada nnUNet, para efetuar a segmentac ¸ão semântica.…”

Section: Trabalhos Relacionadosunclassified

“…Por outro lado, em [Golts et al 2022], implementa uma cascada baseada nas redes 3D U-Net aplicada em dois estágios, que buscam informac ¸ões contextuais espaciais em baixa e alta resoluc ¸ões, respectivamente. e o segundo é aplicado em dados de entrada de alta resoluc ¸ão pela segmentac ¸ão dos rins e tumores renais.…”

Section: Trabalhos Relacionadosunclassified

PPM-DeepLab: Módulo de Pirâmide de Pooling como Codificador da rede DeepLabV3+ para Segmentação de Rins, Cistos e Tumores Renais

Matos,

Oliveira,

Diniz

et al. 2023

Anais Do XXIII Simpósio Brasileiro De Computação Aplicada À Saúde (SBCAS 2023)

View full text Add to dashboard Cite

O câncer renal está entre as principais causas de mortes relacionadas ao câncer em todo o mundo. A detecção e o diagnóstico precoces são cruciais na luta contra esta doença. Recentemente, redes neurais convolucionais (CNNs) demonstraram sua eficácia em tarefas de segmentação semântica. Neste estudo, apresentamos o PPM-DeepLab, um novo modelo arquitetônico projetado para a segmentação de rins, cistos e tumores em imagens de tomografia computadorizada (TC). Especificamente, exploramos o Pyramid Pooling Module (PPM) para aprimorar a rede DeepLabv3+ incorporando informações contextuais de várias escalas. Nosso modelo proposto alcançou resultados promissores, com índices de Dice de 94,89% para rins, 83,95% para cistos e 84,62% para tumores renais.

show abstract

An Ensemble of 3D U-Net Based Models for Segmentation of Kidney and Masses in CT Scans

Cited by 7 publications

References 14 publications

Effect of Dataset Size and Medical Image Modality on Convolutional Neural Network Model Performance for Automated Segmentation: A CT and MR Renal Tumor Imaging Study

Effect of Dataset Size and Medical Image Modality on Convolutional Neural Network Model Performance for Automated Segmentation: A CT and MR Renal Tumor Imaging Study

FuseMedML: a framework for accelerated discovery in machine learning based biomedicine

PPM-DeepLab: Módulo de Pirâmide de Pooling como Codificador da rede DeepLabV3+ para Segmentação de Rins, Cistos e Tumores Renais

Contact Info

Product

Resources

About