The authors use observed features of white, black, and gray fibers in muscle cross-sections to classify the degree of abnormality of the muscle. Muscle fiber patterns are represented as graphs. The authors define a distance function that indicates the degree of deviation of the graph representing a given muscle sample from a reference (normal) graph and use it to classify the samples. The feature extraction and segmentation are handled manually; only the classification is automatic. The distance function is based on a measure of the transformations required to convert the sample graph to the reference graph. In addition, the distance function is used to compute the features of a linear and a K-nearest neighbor classifier. The paper gives experimental results for 211 samples.
This paper is intended to illustrate the applicability of graph theory to a diagnostic problem in medicine. The results indicate some promise for the technique, but the fraction of misclassifications is fairly high. Furthermore, the weights assigned to different elements of the distance function are selected “heuristically”; the difficulty of this trial-and-error process is unclear.
Unfortunately, the paper is nearly unreadable due to its poor organization and the lack of editing for English grammar (e.g., “Both doctors have different opinions of the sample”). The paper was apparently written by the first author (who is from Spain). Figure 1 shows 35 photographs of “typical muscle tissue patterns” with absolutely no caption or text explanation of their meaning. Table 1 contains the results of manual classification of 245 samples by two physicians. The table cannot, however, be read until the classifications are explained, 13 pages later. Even then, there are strange mysteries here. The text states that there is only one sample of Class 5; I found two in the table. The results section mentions 211 samples, half used for training; page 22 states that there were 74 training samples. Unless one is specifically interested in classifying muscle tissue, these inconsistencies, the poor syntax, and the organization of the paper mean that the effort required to read it is too great.