notesum.ai

Published at December 3

U-Net in Medical Image Segmentation: A Review of Its Applications Across Modalities

eess.IV

cs.AI

cs.CV

cs.LG

Released Date: December 3, 2024

Authors: Fnu Neha¹, Deepshikha Bhati, Deepak Kumar Shukla, Sonavi Makarand Dalvi, Nikolaos Mantzou, Safa Shubbar

Aff.: ¹Kent State University, Kent, Ohio, USA

Arxiv: http://arxiv.org/pdf/2412.02242v1

Refer to caption

Study	Modality	Focus Area	Methodology	Performance Metrics
Deng et al. (2024) [54]	X-ray (Anterior-Posterior and Lateral)	Vertebrae instance segmentation for spinal disorder diagnosis	Enhanced U-Net architecture using ConvNeXt as encoder, Informational feature enhancement (IFE) module for texture and edge enhancement, attention in bottleneck, and Residual Network (ResNet) blocks in decoder.	Accuracy: 88.0%, Dice Similarity Coefficient (DSC): 90.6%, Mean intersection over Union (IoU): 79.3%
Haennah et al. (2024) [55]	Chest X-ray (CXR)	Early diagnosis of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)	Classification with Fused U-Net Convolutional Neural Network (FUCNN) optimized by Chaotic System-based Moth Flame Optimization (CSMFO)	Accuracy: 98.5%, Sensitivity: 98.6%, Specificity: 98.9%, Precision: 98.9%
Sharma et al. (2024) [56]	Chest X-ray (CXR)	Tuberculosis Detection	U-Net for lung segmentation, Xception for classification, with gradient-weighted class activation mapping (Grad-CAM) for visualization	Segmentation: Accuracy: 96.5%, Jaccard Index: 90.4%, Dice Coefficient Index (DCI): 94.8%, Classification: Accuracy: 99.3%, Precision: 99.3%, Recall: 99.3%, F1-score: 99.3%
Ying et al. (2024) [57]	Clinical X-ray	Dental Caries Detection	Comparison of object detection: You Only Look Once version 5 (YOLOv5), Detection Transformer (DETR), and segmentation networks: U-Net, and transformer-based U-Net	F1-score: YOLOv5 (87.0%), DETR (82.0%); U-Net (80.0%); transformer-based U-Net (86.0%)
Budagam et al. (2024) [58]	Dental X-ray (Panoramic)	Teeth segmentation and recognition	U-Net and YOLO version 8 (BB-UNet)	mean average precision (mAP): 72.9%, Precision: 94.3%, Recall: 92.3%, DCI: 84.0%
Lyu et al. (2023) [59]	Chest X-ray	Lung and heart segmentation	Multiple tasking Wasserstein generative adversarial network U-Net	Dice Similarity: 95.3%, Precision: 96.4%, F1-score: 95.9%, IoU (Lung): 81.4%, IoU (Heart): 74.6%, DCI (Lung): 85.2%, DCI (Heart): 71.2%
Wu et al. (2021) [60]	Chest X-ray (CXR)	COVID-19 Detection	Modified U-Net-based CNN model for binary classification (COVID-19 vs. Normal) and multiclass classification (COVID-19 vs. Normal vs. Viral Pneumonia)	Binary classification: Accuracy: 99.5%; Multiclass classification: Accuracy: 95.4%
Mosquera-Berrazueta et al. (2023) [61]	Chest X-ray (CXR)	Tuberculosis lesion detection and segmentation	an optimized U-net variant, using ten-fold stratified cross-validation	DCI: 92.0%, IoU: 86.0%
Agarwal et al. (2023) [62]	Chest X-ray (CXR) and CT-scan	Lung segmentation	Proposed a UNet-based model incorporating residual learning and attention mechanisms;	average DCI: 96.4%; Average Jaccard Index (JI): 93.1%
Kholiavchenko et al. (2020) [63]	Chest X-ray (CXR)	Organ segmentation (lung fields, heart, clavicles)	Augmented state-of-the-art CNNs (UNet, LinkNet with ResNeXt, Tiramisu with DenseNet) with organ contour information;	JI: 97.1% (lung fields), 93.3% (heart), 90.3% (clavicles)