Feature-driven breast cancer classification via hybrid model using mammogram images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Deep learning shows promise in medical imaging for breast cancer (BC) diagnosis.
• Traditional machine learning (ML) methods require time-consuming feature extraction.
• Accurate BC detection aids radiologists in timely diagnosis and treatment planning.

Purpose of the Study:

• To propose a novel Feature-driven Breast Cancer Classification (F-BCC-ML) model using a Modified Loss and Activation function-assisted LeNet (MLAL).
• To enhance BC detection accuracy by integrating advanced image processing and deep learning techniques.
• To reduce reliance on manual feature engineering in ML-based BC diagnosis.

Main Methods:

• Mammogram images were enhanced using Improved Bilateral Filtering Technique (IBFT) for noise reduction.
• Image segmentation was performed using SegNet, a deep learning model.
• Feature extraction included Weber Local descriptor assisted Local Gabor XOR Pattern (WLD-LGXP), Median Binary Pattern (MBP), color, and deep features.
• Classification was achieved using the MLAL model combined with a Deep Convolutional Neural Network (DCNN).

Main Results:

• The MLAL+DCNN model achieved a maximum accuracy of 0.936.
• The model demonstrated high precision, reaching 0.947.
• An F-measure of 0.942 was recorded, indicating robust classification performance.

Conclusions:

• The proposed F-BCC-ML model effectively classifies breast cancer from mammograms.
• The integration of IBFT, SegNet, and MLAL+DCNN offers a powerful tool for BC detection.
• This approach supports radiologists by providing accurate and efficient diagnostic predictions.