Table 1 Results of various DL workloads using GaNDLF for multiple anatomies.
From: GaNDLF: the generally nuanced deep learning framework for scalable end-to-end clinical workflows
Task | Organ | Application | Dims | Input modalities (number): type | Output classes | Architecture | Metric | |
|---|---|---|---|---|---|---|---|---|
Type | Average value | |||||||
Segmentation | Brain | Brain extraction | 3 | (1): T1, T1Gd, T2, T2-FLAIR as individual inputs | 1 | UNet | Dice | 0.97 ± 0.01 |
ResUNet | Dice | 0.98 ± 0.01 | ||||||
FCN | Dice | 0.97 ± 0.01 | ||||||
Tumor sub-region segmentation | (4): T1, T1Gd, T2, T2-FLAIR | 3 | UNet | Dice | 0.65 ± 0.05 | |||
ResUNet | Dice | 0.71 ± 0.05 | ||||||
FCN | Dice | 0.62 ± 0.05 | ||||||
UInc | Dice | 0.64 ± 0.05 | ||||||
Brain parcellation | (1): T1 | 133 | ResUNet | Dice | 0.68 ± 0.15 | |||
UNet | Dice | 0.57 ± 0.26 | ||||||
Breast | Breast segmentation | 3 | (1): Digital breast tomosynthesis | 3 | UNet | Dice | 0.78 ± 0.09 | |
Tumor segmentation | (3): T1 pre, peak, and post-contrast injection | 1 | ResUNet | Dice | 0.74 ± 0.01 | |||
Lung | Lung field segmentation | 3 | (1): CT [Lung Cancer Screening] | 1 | ResUNet | Dice | 0.95 ± 0.02 | |
(1): CT [COVID-19] | ResUNet | Dice | 0.97 ± 0.01 | |||||
Eye | Fundus segmentation | 2 | (1): RGB Fundus Images | 1 | UNet | Dice | 0.85 ± 0.04 | |
ResUNet | Dice | 0.90 ± 0.05 | ||||||
FCN | Dice | 0.81 ± 0.04 | ||||||
UInc | Dice | 0.83 ± 0.03 | ||||||
Dental | Quadrant segmentation | 2 | (1): X-Ray | 4 | UNet | Dice | 0.91 ± 0.01 | |
ResUNet | Dice | 0.88 ± 0.01 | ||||||
FCN | Dice | 0.85 ± 0.02 | ||||||
Colon | Colorectal cancer segmentation | 2 | (1): Histology H&E | 1 | ResUNet | Dice | 0.78 ± 0.03 | |
Regression | Brain | Age prediction | 2 | (1): T1 slices | 1 | Specialized VGG | MSE | 0.0141 ± 0.01 |
Classification | Brain | EGFRvIII status prediction | 3 | (4): T1, T1Gd, T2, T2-FLAIR | 2 | VGG11 | Acc | 0.74 ± 0.08 |
Foot | Diabetic foot ulceration | 2 | (1): RGB Foot Images | 4 | VGG11 | Acc | 0.92 ± 0.01 | |
VGG16 | Acc | 0.90 ± 0.01 | ||||||
VGG19 | Acc | 0.89 ± 0.01 | ||||||
DenseNet121 | Acc | 0.87 ± 0.01 | ||||||
Pan-Cancer | TIL Prediction | 2 | (1): Histology H&E | 2 | ImageNet_ VGG16 | Acc | 0.89 ± 0.01 | |
