Classification of Histological Lung Images Using Texture and Morphometric Feature Fusion
Annotatsiya
This study proposes a hybrid method that combines Histogram of Oriented Gradients (HOG) with morphometric features - such as cell area and perimeter - to enhance the performance of image classification algorithms. Multiple machine learning classifiers, including Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Decision Tree (DT), Random Forest (RF), and Naive Bayes (NB), were evaluated on a dataset of annotated lung histological images. Experimental results demonstrate that the integration of texture and geometric features significantly improves classification accuracy. The Random Forest classifier achieved the highest accuracy of 81.90% when using the fused feature set, compared to 73.24% using only texture features. Furthermore, the custom neural network achieved an accuracy of 87.67% on the validation set. These findings confirm that feature fusion enables a more comprehensive representation of histological patterns and supports the development of robust diagnostic systems in digital pathology.