All Projects
Medical AIComputer VisionFeatured

Breast Cancer Ultrasound Segmentation

9-architecture segmentation benchmark on 780 BUSI images. DeepLabV3+ tops with Dice 0.7863, IoU 0.6483. FCN → SimpleUNet → SegNet → Attention-UNet → TransUNet → ResNet34-UNet → EfficientNet-UNet → DeepLabV3+ → Swin-UNet.

0.7863
Best Dice (DeepLabV3+)
0.6483
Best IoU
9
Architectures tested
780 images
Dataset size
Dataset

780 breast ultrasound images (437 benign, 210 malignant, 133 normal)

Approach

9-architecture benchmark: classical encoders → transformer (TransUNet) → ASPP (DeepLabV3+)

Tech Stack
PythonPyTorchsegmentation_models_pytorchtimmAlbumentationsCUDA Tesla T4
Keywords
U-NetDeepLabV3+ASPPSegmentationPyTorchMedical Imaging
Visualizations7 Charts
Deep Dive

Comprehensive benchmark of 9 segmentation architectures on the Breast Ultrasound Images (BUSI) dataset.

Dataset

  • 780 images: 437 benign / 210 malignant / 133 normal (class-stratified)
  • Split: 585 train / 117 val / 78 test
  • Preprocessing: 256×256 resize, Albumentations augmentation (elastic, brightness/contrast, horizontal flip)
  • Loss: 50/50 BCE + Dice combined

Full Architecture Benchmark

ModelParamsDiceIoUType
FCN0.4822Semantic baseline
SimpleUNet31M0.70300.5424Standard encoder-decoder
SegNet7M0.6725Pooling-index decoder
Attention UNet31M0.6964Attention gates on skip
TransUNet124M0.6663ViT encoder + CNN decoder
EfficientNet-UNet20M0.72410.5683EfficientNet-B4 encoder
ResNet34-UNet24M0.78370.6440Pretrained ResNet encoder
DeepLabV3+27M0.78630.6483ASPP multi-scale
Swin-UNetHighTransformer-based

Training Setup

  • 30 epochs, AdamW + cosine annealing LR
  • Mixed precision (AMP), early stopping
  • Metrics: Dice, IoU, F1, Precision, Recall (TorchMetrics)

Why DeepLabV3+ Wins Atrous Spatial Pyramid Pooling (ASPP) with output_stride=16 extracts multi-scale context without spatial collapse — critical for breast tumors that range from 2mm to 30mm. This prevents the spatial resolution loss that hurts standard encoders on small lesions.