Scribd
Upload
4 views2 pages

Encoder Decoder Breast Cancer Project Description

This project introduces a U-Net–style encoder–decoder deep learning model for the automatic segmentation and classification of breast cancer images, aimed at improving computer-assisted diagnosis. The model effectively detects cancerous regions in mammograms and histopathological slides by utilizing a convolutional neural network architecture with features like skip connections and batch normalization. It is designed for integration into diagnostic workflows and supports applications in automated tumor detection, medical image annotation, and research.

Uploaded by

Makram Ettriki
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views2 pages

Encoder Decoder Breast Cancer Project Description

This project introduces a U-Net–style encoder–decoder deep learning model for the automatic segmentation and classification of breast cancer images, aimed at improving computer-assisted diagnosis. The model effectively detects cancerous regions in mammograms and histopathological slides by utilizing a convolutional neural network architecture with features like skip connections and batch normalization. It is designed for integration into diagnostic workflows and supports applications in automated tumor detection, medical image annotation, and research.

Uploaded by

Makram Ettriki
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Encoder–Decoder Model for the Segmentation and

Classification of Breast Cancer Images

1. Project Overview
This project presents a deep learning model based on a U-Net–style encoder–decoder
architecture for the automatic segmentation and classification of breast cancer images.
The model is designed to detect and delineate cancerous regions within breast tissue
images, such as mammograms or histopathological slides, enabling more accurate
computer-assisted diagnosis.

2. Objectives
- Build a convolutional neural network capable of pixel-wise segmentation of breast cancer
lesions.
- Extract meaningful spatial and contextual features for distinguishing between cancerous
and non-cancerous regions.
- Develop a robust and generalizable architecture that can be integrated into diagnostic or
research workflows.

3. Model Architecture
The model follows a U-Net–based encoder–decoder design implemented in
TensorFlow/Keras. Encoder (Contracting Path):
- Multiple EncoderMiniBlock units with 3×3 convolutions, ReLU activations, and He
initialization.
- Batch Normalization and Dropout enhance generalization.
- MaxPooling reduces spatial resolution while capturing deeper contextual features.

Decoder (Expanding Path):


- DecoderMiniBlock units use Conv2DTranspose layers for upsampling.
- Skip connections concatenate encoder and decoder features to recover fine spatial
details.

Output Layer:
- A 1×1 convolution layer with Softmax activation produces a segmentation mask with two
classes (tumor and normal tissue).

4. Implementation Details
- Framework: TensorFlow / Keras
- Input Size: 640 × 640 × 3
- Base Filters: 32 (doubles at each encoder level)
- Activations: ReLU (convolution layers), Softmax (output)
- Loss Function: Categorical cross-entropy or Dice loss
- Optimizer: Adam optimizer with adjustable learning rate

5. Dataset and Preprocessing


- Designed for datasets containing breast cancer images such as CBIS-DDSM, BreakHis,
or clinical datasets.
- Input images are resized and normalized.
- Ground truth segmentation masks are used for supervised training.

6. Key Advantages
- Accurate tumor boundary detection using skip connections.
- Efficient training with dropout and batch normalization.
- Flexible architecture adaptable for other medical segmentation tasks.
- Supports both segmentation and classification in a single model.

7. Potential Applications
- Breast cancer diagnosis support — automated tumor detection and localization.
- Medical image annotation — assisting radiologists with region labeling.
- Research — feature extraction and data preprocessing for diagnostic AI models.

You might also like