Related Experiment Video
Updated: Jun 4, 2025
08:32
Using Computer-based Image Analysis to Improve Quantification of Lung Metastasis in the 4T1 Breast Cancer Model
Published on: October 2, 2020
6.2K
AI-Driven Microscopy: Cutting-Edge Approach for Breast Tissue Prognosis Using Microscopic Images
Tariq Mahmood1,2, Tanzila Saba1, Shaha Al-Otaibi3
1Artificial Intelligence and Data Analytics (AIDA) lab, CCIS Prince Sultan University, Riyadh, Saudi Arabia.
Microscopy Research and Technique
|January 3, 2025
Summary
This study introduces a deep learning framework for analyzing microscopic breast cancer images, achieving high accuracy in classifying benign/malignant tissues and subtypes. The computer-aided analysis accelerates diagnosis and reduces manual evaluation burdens.
Area of Science:
- Computational pathology
- Medical image analysis
- Deep learning in oncology
Background:
- Microscopic imaging is crucial for disease diagnosis, but manual quantitative evaluation of high-resolution images is challenging.
- Accurate and rapid analysis of breast cancer tissues is essential for clinical diagnosis, prognosis, and treatment.
- Existing methods face difficulties in quantifying intricate tissues and meeting real-time pathological image analysis demands.
Purpose of the Study:
- To develop advanced computer-aided analysis methods for rapid and precise quantitative evaluation of microscopic breast cancer images.
- To enhance the accuracy and efficiency of clinical diagnosis, course analysis, and prognostic prediction using deep learning.
- To address the challenges of quantifying small, intricate breast cancer tissues and meeting real-time pathological image analysis requirements.
Main Methods:
- Integration of a dense convolutional network (DenseNet) with an attention mechanism for enhanced feature extraction.
- Utilizing deep learning frameworks like squeeze-and-excitation and dilated dense convolution blocks.
- Employing transfer learning with pre-trained architectures (VGGNet-19, ResNet152V2, EfficientNetV2-B1, DenseNet-121) and data enhancement techniques.
Main Results:
- Achieved 99.6% accuracy for secondary classification (benign/malignant) and 99.4% for eight breast subtypes classification.
- Demonstrated substantial improvement over existing methods, which typically report lower accuracies (85%-94%) for breast subtype classification.
- The framework effectively performs precise prediction and segmentation of breast lesions using multi-scale feature extraction and attention mechanisms.
Conclusions:
- The proposed deep learning framework offers a sophisticated and efficient solution for quantitative pathological image analysis.
- This approach significantly improves diagnostic accuracy and reduces the time and cost associated with traditional methods.
- The high accuracy achieved provides reliable diagnostic support, enhancing precision in clinical decision-making for breast cancer.
