Comparing 3D, 2.5D, and 2D Approaches to Brain Image Segmentation

Avesta, Arman; Hossain, S. M. Khaled; Lin, MingDe; Aboian, Mariam; Krumholz, Harlan M.; Aneja, Sanjay

doi:10.1101/2022.11.03.22281923

Cited by 4 publications

(3 citation statements)

References 40 publications

(81 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Firstly, the increase in the number of parameters of the 3D network leads to an increase in computational complexity and training time, and requires higher computing power. Research findings indicate that The 3D approach requires 20 times more computational memory than 2D approaches (Avesta et al 2022). Secondly, with limited GPU memory, 2D networks can efficiently capture broad and diverse receptive fields, enabling the aggregation of in-plane, multiscale contextual cues for satisfactory segmentation (Bian et al 2018).…”

Section: Choice Of 2d and 3d Networkmentioning

confidence: 99%

Multi-scale adversarial learning with difficult region supervision learning models for primary tumor segmentation

Zheng,

Sun,

et al. 2024

Phys. Med. Biol.

View full text Add to dashboard Cite

Objective. Recently, deep learning techniques have found extensive application in accurate and automated segmentation of tumor regions. However, owing to the variety of tumor shapes, complex types, and unpredictability of spatial distribution, tumor segmentation still faces major challenges. Taking cues from the deep supervision and adversarial learning, we have devised a cascade-based methodology incorporating multi-scale adversarial learning and difficult-region supervision learning in this study to tackle these challenges.  Approach. Overall, the method adheres to a coarse-to-fine strategy, first roughly locating the target region, and then refining the target object with multi-stage cascaded binary segmentation which converts complex multi-class segmentation problems into multiple simpler binary segmentation problems. In addition, a Multi-Scale Adversarial Learning Difficult Supervised UNet (MSALDS-UNet) is proposed as our model for fine-segmentation, which applies multiple discriminators along the decoding path of the segmentation network to implement multi-scale adversarial learning, thereby enhancing the accuracy of network segmentation. Meanwhile, in MSALDS-UNet, we introduce a difficult region supervision loss to effectively utilize structural information for segmenting difficult-to-distinguish areas, such as blurry boundary areas. Main results. A thorough validation of three independent public databases (KiTS21, MSD's Brain and Pancreas datasets) shows that our model achieves satisfactory results for tumor segmentation in terms of key evaluation metrics including \textcolor{red}{DSC, JC, and HD95}.Significance. This paper introduces a cascade approach that combines multi-scale adversarial learning and difficult supervision to achieve precise tumor segmentation. It confirms that the combination can improve the segmentation performance, especially for small objects. (Our code will be publicly available after acceptance on https://zhengshenhai.github.io/)

show abstract

Section: Choice Of 2d and 3d Networkmentioning

confidence: 99%

Multi-scale adversarial learning with difficult region supervision learning models for primary tumor segmentation

Zheng,

Sun,

et al. 2024

Phys. Med. Biol.

View full text Add to dashboard Cite

show abstract

“…The model was trained utilizing the computational power of an NVIDIA Tesla V100 GPU. Although some studies favor 2.5D and 3D U-Nets over 2D U-Nets (23,24), providing this extra information doesn't consistently enhance accuracy (15). Additionally, 2D CNNs are more computationally efficient than 2.5D or 3D U-Nets, requiring fewer resources for processing.…”

Section: In-house Model For Automated Delineationmentioning

confidence: 99%

Clinical acceptance and dosimetric impact of automatically delineated elective target and organs at risk for head and neck MR-Linac patients

Koteva,

Eiben,

Dunlop

et al. 2024

Front. Oncol.

View full text Add to dashboard Cite

BackgroundMR-Linac allows for daily online treatment adaptation to the observed geometry of tumor targets and organs at risk (OARs). Manual delineation for head and neck cancer (HNC) patients takes 45-75 minutes, making it unsuitable for online adaptive radiotherapy. This study aims to clinically and dosimetrically validate an in-house developed algorithm which automatically delineates the elective target volume and OARs for HNC patients in under a minute.MethodsAuto-contours were generated by an in-house model with 2D U-Net architecture trained and tested on 52 MRI scans via leave-one-out cross-validation. A randomized selection of 684 automated and manual contours (split half-and-half) was presented to an oncologist to perform a blind test and determine the clinical acceptability. The dosimetric impact was investigated for 13 patients evaluating the differences in dosage for all structures.ResultsAutomated contours were generated in 8 seconds per MRI scan. The blind test concluded that 114 (33%) of auto-contours required adjustments with 85 only minor and 15 (4.4%) of manual contours required adjustments with 12 only minor. Dosimetric analysis showed negligible dosimetric differences between clinically acceptable structures and structures requiring minor changes. The Dice Similarity coefficients for the auto-contours ranged from 0.66 ± 0.11 to 0.88 ± 0.06 across all structures.ConclusionMajority of auto-contours were clinically acceptable and could be used without any adjustments. Majority of structures requiring minor adjustments did not lead to significant dosimetric differences, hence manual adjustments were needed only for structures requiring major changes, which takes no longer than 10 minutes per patient.

show abstract

“…They discovered that the 3D brain MRIs far outperformed the 2D and 2.5D inputs. However, the 3D inputs required more memory for training (11).…”

Section: Related Workmentioning

confidence: 99%

Brain Tumor Segmentation Based on Deep Learning, Attention Mechanisms, and Energy-Based Uncertainty Prediction

Schwehr,

Achanta

2023

Preprint

View full text Add to dashboard Cite

<p>Brain tumors are one of the deadliest forms of cancer with a mortality rate of over 80%. A quick and accurate diagnosis is crucial to increase the chance of survival. However, in medical analysis, the manual annotation and segmentation of a brain tumor can be a complicated task. Multiple MRI modalities are typically analyzed as they provide unique information regarding the tumor regions. Although these MRI modalities are helpful for segmenting gliomas, they tend to increase overfitting and computation. This paper proposes a region of interest detection algorithm that is implemented during data preprocessing to locate salient features and remove extraneous MRI data. This decreases the input size, allowing for more aggressive data augmentations and deeper neural networks. Following the preprocessing of the MRI modalities, a fully convolutional autoencoder with soft attention segments the different brain MRIs. When these deep learning algorithms are implemented in practice, analysts and physicians cannot differentiate between accurate and inaccurate predictions. Subsequently, test time augmentations and an energy-based model were used for voxel-based uncertainty predictions. Experimentation was conducted on the BraTS benchmarks and achieved state-of-the-art segmentation performance. Additionally, qualitative results were used to assess the segmentation models and uncertainty predictions. The code for this work is made available online at: https://github.com/WeToTheMoon/BrainTumorSegmentation.</p>

show abstract

Comparing 3D, 2.5D, and 2D Approaches to Brain Image Segmentation

Cited by 4 publications

References 40 publications

Multi-scale adversarial learning with difficult region supervision learning models for primary tumor segmentation

Multi-scale adversarial learning with difficult region supervision learning models for primary tumor segmentation

Clinical acceptance and dosimetric impact of automatically delineated elective target and organs at risk for head and neck MR-Linac patients

Brain Tumor Segmentation Based on Deep Learning, Attention Mechanisms, and Energy-Based Uncertainty Prediction

Contact Info

Product

Resources

About