Interstitial Lung Disease diagnosis relies on Computer Tomography analysis, which is challenging due to overlapping radiological findings (indicating the stage and type of lung damage) and patient anatomical variability, demanding extensive expert effort. Machine learning, particularly supervised models like convolutional neural networks, has been applied to automate this analysis, but its reliance on fully annotated data limits widespread use. To address this, unsupervised methods based on clustering algorithms are being developed. This work proposes an unsupervised method combining superpixel extraction, characterization, and DBSCAN clustering for fibrosis segmentation. By preserving object boundaries through superpixels and incorporating first- and second-order features, the method achieves an accuracy of 0.6714, surpassing state-of-the-art approaches and showing promise for effective fibrosis segmentation.