Wavelet Transform and Texton based Analysis for Detection of Benign and Malignant Masses

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Cancer's low survival rate necessitates early diagnosis.
• Initial cancer stages are often asymptomatic, increasing reliance on screening.
• Computer-aided diagnostic models enhance diagnostic efficiency.

Purpose of the Study:

• To propose a novel technique for differentiating benign and malignant breast lesions in mammograms.
• To improve the accuracy and efficiency of breast cancer diagnosis using digital mammography.

Main Methods:

• Utilized 400 mammograms from the Digital Database for Screening Mammography.
• Applied multiresolution analysis (Haar wavelet) and Schmid Filter Bank for lesion analysis.
• Extracted statistical and Haralick's features, followed by Partition Membership Filter for feature partitioning.

Main Results:

• Achieved a maximum accuracy of 98.63% in differentiating lesions.
• Obtained an Area Under the Curve (AUC) of 0.981.
• Validated the model using Random Forest Classifier and ten-fold cross-validation.

Conclusions:

• The proposed technique demonstrates high accuracy in classifying breast lesions.
• This novel approach shows significant potential for enhancing early breast cancer detection.
• The method offers a promising tool for computer-aided diagnosis in mammography.