K-Means Porosity Clustering in MRI-Scanned Reef Rock: Insights from Zechstein Limestone, Poland

Abstract Objectives/scope The study aimed to segment the pore space in a 3D MRI scan of a reef limestone using k-means clustering. Voxels with similar proton densities were grouped into 3D clusters. Traditional MRI interpretation struggles with highly heterogeneous samples, as automatic color mapping can create an excessive number of value ranges, reducing the readability of an image. Unlike manual segmentation, this workflow simplifies the image while retaining essential structural details of the pore space. Methods, procedures, process A brine-saturated bioclastic limestone sample from the Zechstein Limestone Formation (Ca1, Poland) was subjected to MRI scanning at 30 MHz. The 3D SE-SPI sequence with echo time of 2000 μs, repetition time of 10000 ms and voxel size of 0.625 mm was used. The image consisting of mL-scaled voxels was segmented using k-means clustering. The number of clusters was determined using the elbow method, resulting in four clusters. The first proton density cluster containing noise was excluded. The remaining three clusters had their volumes computed along with the mean, standard deviation and median of the voxel values they contained. Results, observations, conclusions The analysis showed that the three identified pore domains characterized by low, moderate and high proton density, had volumes ranging from around 0.7 mL in the low and medium proton density clusters to roughly 0.4 mL in the high proton density cluster. The mean voxel values showed an increase of at least 20% between successive clusters of growing proton density, while the standard deviation was the highest in the cluster with the greatest proton density. The segmented image revealed multiple bendy, channel-shaped dissolution features, containing varying amounts of water. In places where dissolution had intensified, larger pore domains were created. The k-means clustering proved to be a robust method for distinguishing similar pore groups in MRI data, allowing for their consistent and reproducible quantification. Novel/additive information Compared to medical studies, the k-means MRI data clustering remains uncommon in petrophysics. We facilitate MRI-based pore structure analysis in carbonate rocks by offering a data-driven alternative to manual segmentation. This unbiased approach enables consistent classification of porosity. Unlike traditional color mapping, which can obscure critical details in complex pore networks, clustering highlights meaningful variations in proton density. The proposed methodology enhances porosity assessment and provides insights into permeability by identifying interconnected pore pathways.