Improved Algorithm for Image Segmentation based on the Three-dimensional Reconstruction of Tumor Images

Jianning Han, Quan Zhang, Peng Yang, Yifan GongInternational Journal of Signal Processing Image Processing and Pattern Recognition 2015 · DOI

Abstract: Recently, more and more researchers have been paying close attention to tumor image processing techniques, in which the tumor image segmentation technology is regarded as the absolute research focus. Given that there exist only one tumor image segmentation algorithm, no further high quality three-dimensional slice can be provided for the later reconstruction of the three-dimensional reconstruction of tumor images. In this paper, we propose an improved segmentation algorithm by combining the fuzzy clustering segmentation with the fuzzy edge enhancement. Finally, the experimental results show that the fuzzy clustering segmentation is highly efficient for dealing with the fuzziness and complexity of brain tissues, and the images after the fuzzy clustering segmentation calculated by the edge detection algorithm of fuzzy enhancement have more clear edges, which provides a solid foundation for further 3D processing and finally helps achieve better 3D visualization effect on brain tumors. Our algorithm provides promising application prospect in the field of the medical image processing.

What Problem Does This Paper Address?

Accurate brain tumor segmentation is the "bottleneck" of medical imaging. Traditional methods, such as the watershed algorithm, are notoriously sensitive to noise, often leading to "over-segmentation" where the image is cluttered with meaningless "catchment basins." Furthermore, standard edge detection operators (like Canny or Prewitt) struggle with the low contrast of MRI and CT scans.

Original brain tumor image showing the low contrast and fuzzy boundaries typical of medical scans.

In these medical images, the transition between healthy tissue and a tumor isn't a sharp line; it’s a "fuzzy" gradient. Traditional algorithms often lose critical internal details or produce disconnected edges, which makes high-quality 3D reconstruction—essential for surgical planning and radiation therapy—nearly impossible.

Methodology

The authors propose a two-stage "Comprehensive Improvement" algorithm designed to handle the inherent uncertainty of biological tissue:

  1. Fuzzy C-Means (FCM) Clustering: Instead of assigning a pixel to just one category (e.g., "tumor" or "not tumor"), FCM assigns a degree of membership to multiple clusters based on gray level, texture, and statistical parameters. This better represents the overlapping characteristics of human tissue.
  2. Fuzzy Edge Enhancement: After clustering, the authors apply a specialized edge detection pipeline. This includes using Otsu’s method for thresholding and a membership function adjustment to enhance contrast in both low and high gray-level regions. They utilize the Nakagowa and Rosenfeld "min" operator to extract edges, preventing the loss of information that typically occurs when ordinary operators "cut to zero" in low-value areas.

The final 3D reconstruction was implemented using VC++ and the Visualization Toolkit (VTK) to transform these processed 2D slices into an interactive 3D model.

Key Insights and Ideas

  • Embracing Uncertainty: The core innovation is the recognition that medical images are inherently fuzzy. By using fuzzy logic, the algorithm preserves the "marginal information" that traditional binary logic discards.
  • Feedback Control: The authors added a feedback loop to the clustering process to refine object choice and similarity measures, ensuring the algorithm adapts to the specific complexity of brain tissue.
  • Solving the Connectivity Problem: Standard edge detectors often produce "broken" lines. The fuzzy enhancement stage specifically targets the connectivity of internal structures, ensuring that the internal configuration of the tumor is visible, not just its outer shell.

Key Results

The experimental results confirmed that FCM is highly efficient at dealing with the inherent fuzziness of brain tissues compared to traditional structural or statistical methods. Compared to the watershed algorithm (which left internal details unclear) and standard operators (which caused internal boundaries to disappear), the proposed method produced distinctive external edges and clear internal structures.

The result of the proposed fuzzy clustering and edge enhancement algorithm, showing clear tissue differentiation and boundary connectivity.

This qualitative superiority provides a "scientific basis" for doctors to see the actual volume and shape of a tumor in relation to the rest of the brain. The authors successfully generated 3D models where the tumor could be extracted and viewed independently, offering a promising tool for clinical diagnosis and surgical preparation.

Limitations and Future Work

While the visual results are compelling, the paper relies heavily on qualitative comparisons. It lacks standard quantitative metrics common in modern segmentation papers, such as the Dice Similarity Coefficient or Jaccard Index. Additionally, the computational complexity of fuzzy algorithms and 3D reconstruction is not discussed, which is a critical factor for real-time clinical application.

Future work should focus on moving the algorithm into clinical application, medical education, and real-time surgical assistance, while incorporating more robust quantitative validation.

Summary updated

Abstract

Recently, more and more researchers have been paying close attention to tumor image processing techniques, in which the tumor image segmentation technology is regarded as the absolute research focus. Given that there exist only one tumor image segmentation algorithm, no further high quality three-dimensional slice can be provided for the later reconstruction of the three-dimensional reconstruction of tumor images. In this paper, we propose an improved segmentation algorithm by combining the fuzzy clustering segmentation with the fuzzy edge enhancement. Finally, the experimental results show that the fuzzy clustering segmentation is highly efficient for dealing with the fuzziness and complexity of brain tissues, and the images after the fuzzy clustering segmentation calculated by the edge detection algorithm of fuzzy enhancement have more clear edges, which provides a solid foundation for further 3D processing and finally helps achieve better 3D visualization effect on brain tumors. Our algorithm provides promising application prospect in the field of the medical image processing.
BibTeX
@article{han2015improved,
  author = {Jianning Han and Quan Zhang and Peng Yang and Yifan Gong},
  title = {{Improved Algorithm for Image Segmentation based on the Three-dimensional Reconstruction of Tumor Images}},
  journal = {International Journal of Signal Processing Image Processing and Pattern Recognition},
  year = {2015},
  doi = {10.14257/ijsip.2015.8.6.03},
}
Figure