The representations of sonographic image texture for breast cancer using co-occurrence matrix

Shao Jer Chen, Kuo Sheng Cheng, Yuan Chang Dai, Yung Nien Sun, Yen Ting Chen, Ku Yaw Chang, Wen Ching Hsu, Tsai Wang Chang

Research output: Contribution to journalArticlepeer-review

9 Citations (Scopus)


In this paper, the feature parameters derived from the co-occurrence matrix are applied to quantitatively characterize the texture features of the ultrasound images for breast cancer diagnosis. Further correlation with the histopathological finding is also developed for facilitating clinical diagnosis. Thirty-three patients are recruited for this study. The ultrasound imaging system in clinical use is ATL HDI 3000. The parameters used for image acquisition are kept in the same conditions during clinical examination. The image characteristics of various texture features are demonstrated with examples of various breast nodules. The associated parameters for texture features of the nodules can be summarized as follows: 1. Energy: homogenous texture; 2. Entropy: disorder of the image; 3. Sum average: homogeneously brightness; 4. Contrast: difference of gray-scale through continuous pixels of the image; 5. Variance: heterogeneity; 6. Inverse differential moment (IDM): local homogeneity; and 7. Correlation: linear relationship between the gray-scale of pixel pairs. From the results, it is shown that the breast nodule revealing high energy, high IDM value, low contrast, low variance, low entropy, or low sum average value in its sonographic image suggest the case of breast cancer with the pattern of delicate fibrous stroma and multiple small tumorous foci in pathological findings.

Original languageEnglish
Pages (from-to)193-199
Number of pages7
JournalJournal of Medical and Biological Engineering
Issue number4
Publication statusPublished - 2005 Dec 1

All Science Journal Classification (ASJC) codes

  • Biomedical Engineering


Dive into the research topics of 'The representations of sonographic image texture for breast cancer using co-occurrence matrix'. Together they form a unique fingerprint.

Cite this