Performance Benchmarking and Optimisation of Clustering Algorithms for Local and Non-Local Similarity Measure in Medical Image Analysis

2026-07-10

AI Summary

This study addresses the challenge in medical image post-processing of simultaneously preserving local details and exploiting non-local self-similarity—a balance that traditional global low-rank methods struggle to achieve. The authors present the first systematic evaluation and optimization of five clustering algorithms—k-means, mini-batch k-means, agglomerative hierarchical clustering, BIRCH, and bisecting k-means—across multimodal medical images, including MRI, ultrasound, and chest X-rays. Algorithm performance is assessed using Silhouette, Davies–Bouldin (DB), and Calinski–Harabasz (CH) indices, with hyperparameters tuned via random search. Results reveal that agglomerative clustering achieves the best performance on MRI and ultrasound, while mini-batch k-means offers the most balanced results for X-ray images. Standard k-means and bisecting k-means exhibit high inter-cluster separation but large intra-cluster variation, whereas BIRCH underperforms overall, highlighting a fundamental trade-off between clustering efficiency and the preservation of local image details.

Abstract

Medical imaging generates high-resolution images posing significant storage, transmission, and computational challenges. While low-rank matrix approximation (LoRMA) techniques offer efficient compression by exploiting structural redundancy, global approaches often fail to preserve local details critical for diagnosis. This paper focuses on clustering techniques that exploit non-local self-similarity to identify structurally similar regions in medical images. These clusters can be used for post-processing tasks such as adaptive image compression. We evaluate five clustering techniques: k-means, mini-batch k-means, agglomerative hierarchical clustering, balanced iterative reducing and clustering using hierarchies (BIRCH), and bisecting k-means across MRI, ultrasound, and chest X-ray modalities. All clustering techniques were optimised using random search, and cluster quality was assessed using the Silhouette score, the Davies-Bouldin (DB) index, and the Calinski-Harabasz (CH) index. Results demonstrate that standard k-means and bisecting k-means generally achieve strong cluster cohesion and separation across modalities. However, they tend to form a small number of clusters with high intra-cluster variability, limiting their effectiveness for post-processing tasks such as adaptive compression. Agglomerative clustering outperformed other techniques for MRI and ultrasound in terms of intra-cluster homogeneity, making it more suitable for preserving fine diagnostic details. For chest X-rays, mini-batch k-means achieved the best balance between clustering quality and intra-cluster compactness. BIRCH consistently underperformed across all modalities.

Problem

• clustering algorithms
• non-local similarity
• medical image analysis
• adaptive compression
• local details preservation

Innovation

• clustering algorithms
• non-local self-similarity
• medical image analysis
• adaptive compression
• performance benchmarking