LightGBM: A Leading Force in Breast Cancer Diagnosis Through Machine Learning and Image Processing

Leveraging the power of LightGBM, a leading machine learning model, this approach enhances breast cancer diagnosis through advanced image processing techniques. The process begins with image resizing and augmentation via rotation to increase the dataset's diversity. The first step extracts color histogram features by converting images to the HSV color space, calculating and normalizing the histogram of the HUE channel, resulting in a flattened 1D array. Next, contour features are extracted by converting images to grayscale, finding the largest contours, and calculating properties such as the number of contours, maximum area, and maximum perimeter, yielding another 1D array. Subsequently, HU moments are derived by converting images to grayscale, calculating moments, and computing the seven HU moments, forming a flattened array. Additionally, Haralick features are extracted from grayscale images by calculating the Gray-Level Co-occurrence Matrix (GLCM) and deriving contrast, energy, homogeneity, and correlation metrics, creating a final 1D array. These combined extracted features feed into the LightGBM classifier, which outputs classifications across various breast cancer types, including benign (Adenosis, Fibroadenoma, Phyllodes Tumor, Tubular Adenoma) and malignant (Ductal Carcinoma, Lobular Carcinoma, Mucinous Carcinoma, Papillary Carcinoma) categories, culminating in a high-accuracy diagnosis.

Keywords: Breast Cancer Dataset, Features Extraction, Image Processing Techniques, LightGBM Classifier, Machine Learning Algorithm and Accuracy.