Scribd
Upload
52 views6 pages

AI For Breast Cancer

The document discusses the development of an AI model called MIB-NET for classifying and segmenting breast cancer images to enhance diagnostic accuracy. It highlights the use of machine learning, multitask learning, and interpretable methods to ensure the model's decisions are understandable to healthcare professionals. The model employs advanced image processing techniques and interpretable deep learning methods to improve transparency and trust in AI applications within healthcare.

Uploaded by

24im0002
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
52 views6 pages

AI For Breast Cancer

The document discusses the development of an AI model called MIB-NET for classifying and segmenting breast cancer images to enhance diagnostic accuracy. It highlights the use of machine learning, multitask learning, and interpretable methods to ensure the model's decisions are understandable to healthcare professionals. The model employs advanced image processing techniques and interpretable deep learning methods to improve transparency and trust in AI applications within healthcare.

Uploaded by

24im0002
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
You are on page 1/ 6

AI FOR BREAST CANCER:

INTERPRETABLE MIB-NET
AIM:
The aim of this work is to develop a deep
learning-based model that simultaneously
classifies and segments breast cancer images,
improving diagnosis accuracy.
WHY AI?
AI because it uses machine learning,
deep learning, multitask learning, and
image processing techniques to solve
complex problems in breast cancer
detection and diagnosis.
HOW IT QUALIFIES AS AI:
• Machine Learning (ML): • Multitask Learning:
The model is built using deep learning, a subset of The model performs two tasks simultaneously—
machine learning, which means it learns from large classification and segmentation—using a single
datasets of breast images (e.g., mammograms, AI model.
ultrasounds).
This is a key feature of AI systems designed to
The system "learns" to classify whether an image solve complex problems.
shows a benign or malignant tumor and also
identifies the tumor's location in the image AI systems often leverage multitask learning to
(segmentation). improve efficiency and accuracy.

This is an AI-driven process, as it involves teaching


the system to make decisions based on patterns in
the data.
HOW IT QUALIFIES AS AI:

• Interpretability: • Image Processing:


The use of the Information Bottleneck method The system is designed to analyze and
makes the AI model interpretable, meaning it process medical images (e.g., mammograms,
can explain how it made a decision. ultrasound scans), which is another core
AI models can often be seen as "black boxes," application of AI, particularly in fields like
but the aim of this work is to make the healthcare.
decisions of the model understandable for Deep learning models, like Convolutional
doctors or healthcare professionals. Neural Networks (CNNs), are commonly used
This is important for increasing trust in AI for image recognition tasks, which are central
systems to this thesis.
Interpretable Methods in Deep Learning
Interpretable methods in deep learning are essential for improving the
transparency and trustworthiness of AI models, especially in fields like
healthcare, finance, and autonomous systems.
1. Feature Visualization Method 2. Class Activation Map(CAM):
Creates a heatmap
Deconvolutional highlights relevant areas in the
Network(DeCovNet) input.

3. Gradient Based Method:


1. Saliency Mapping.
2. Guided Backpropagation(GBP)
3. Layerwise Relevance Propagation(LRP)
4. Deep Taylor Decomposition(Deep Taylor)
5. Pattern Attribution
MULTITASK LEARNING
Multitask learning, which considers the relevance of tasks, has been successfully applied
in several applications, such as natural language processing, robotics, computer vision,
biological medicine.

Two Methods used in Multitask Learning:


HARD PARAMETER SHARING SOFT-PARAMETER SHARING
• Definition: The model shares the same hidden layers • Definition: Each task has its own separate model and
(parameters) for all tasks, with separate output layers parameters, but constraints or regularization techniques
for each task.
(e.g., L2 regularization) are applied to encourage the
• Advantages: Reduces the number of parameters, parameters to stay similar.
making the model computationally efficient.
• Advantages: Allows for task-specific flexibility, which is
• Acts as a regularizer, reducing overfitting by forcing
the model to learn generalized features. useful for tasks that are only loosely related.

• Applications: Works well for tasks with a high degree • Applications: Effective for tasks that are related but
of correlation, such as tumor classification and may have different feature distributions or complexities.
segmentation.

You might also like