Table 1.
Literature review on gastrointestinal tract.
Fig 1.
Flow chart for automatic segmentation of small bowel, large bowel, and stomach in GI tract.
Fig 2.
Images of the UW-Madison database.
Fig 3.
Ground Truth Mask Generation (a) Original Image, (b) RLE Encoding for Large Bowel, (c) RLE Encoding for Small Bowel, and (d) RLE Encoding for Stomach.
Table 2.
Dataset splitting in training, testing and validation.
Fig 4.
Sample Images After Applying the Augmentation Techniques; (a) & (e) Original Images, (b) &(f) Horizontal Flip, (c) & (g) Vertical Flip, and (d) & (h) Rotation.
Fig 5.
Block diagram of UMobileNetV2 for segmentation.
Table 3.
Detailed description of layers of the UMobileNet model.
Fig 6.
Loss analysis for different encoders using adam optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 7.
Dice coefficient analysis for different encoders using adam optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 8.
IoU Analysis for Different Encoders using Adam Optimizer (a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 9.
Results comparison of UMobileNet V2 model with different TL models with adam optimizer using test dataset.
Fig 10.
Loss Analysis for Different Encoders using RMS Optimizer (a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 11.
Dice coefficient analysis for different encoders using RMS optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 12.
IoU analysis for different encoders using RMS optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 13.
Results comparison of UMobileNet V2 model with different TL models with RMS optimizer using test dataset.
Fig 14.
Loss analysis for different encoders using SGD optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 15.
Dice coefficient analysis for different encoders using SGD optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 16.
IoU analysis for different encoders using SGD optimizer.
(a) Xception, (b) ResNet 101, (c) NASNet Mobile, and (d) UMobileNet V2 Model.
Fig 17.
Results comparison of UMobileNet V2 model with different TL models at SGD optimizer using test dataset.
Fig 18.
Loss comparison of UMobileNet V2 model with different encoders with different optimizers using test dataset.
Fig 19.
Dice coefficient comparison of UMobileNet V2 model with different encoders with different optimizers using test dataset.
Fig 20.
IoU comparison of UMobileNet V2 model with different encoders with different optimizers using test dataset.
Fig 21.
Graphs of the Optimized Model (a) Dice Coefficient, (b) IoU Coefficient, and (c) Loss.
Fig 22.
Visual examination of the outcomes in the form of pictures.
Here, yellow is used to show the big intestine, green is used to represent the small intestine, and red is used to represent the stomach.
Table 4.
Analysis of optimized model based on different epochs.
Table 5.
Comparison of UMobileNet V2 model with other segmentation models.
Table 6.
State-of-the-art comparison on UW Madison GI tract dataset.