Hejblum G, Jarlier V, Grosset J, Aurengo A
Unité de Méthodologie en Informatique et Statistique en Médecine (Institut National de la Santé et de la Recherche Médicale U-194), Paris, France.
J Clin Microbiol. 1993 Sep;31(9):2396-401. doi: 10.1128/jcm.31.9.2396-2401.1993.
An original algorithm referred to as the radial profile analysis algorithm was implemented on a Macintosh Quadra 700 computer to provide an automatic determination of the inhibition zone diameters of antibiotic susceptibility tests performed with the disk diffusion method. After digitization of the petri plate image, each antibiotic disk is recognized and labeled. Pixels of the local zone around each disk are then used for generating a profile pattern that is subjected to decision rules. The resulting estimate of the inhibition zone diameter is then automatically compared with conventional breakpoints for classifying the tested strain in one of the clinical categories of antibiotic susceptibility. The program is also able to request a human reading for some rare plates difficult to interpret. The algorithm accuracy was tested by comparing the results with a combination of independent human measurements performed on the tested plates. The test sample was composed of 98 strains, and 2,552 tests of 40 distinct antibiotics were subjected to the analysis. The difference between the automatic and human diameter estimates was less than 4 mm in 90% of the tests. The agreement between the automatic and human clinical categorizations amounted to 95.5%, and severe (major and very major) disagreements were found in 5.6% of the tests performed with staphylococci but only 0.3% of the tests with gram-negative rods. We conclude that the radial profile analysis algorithm is a solid backbone for an automatic system dedicated to the clinical interpretation of disk diffusion antibiotic susceptibility tests.
一种名为径向轮廓分析算法的原创算法在Macintosh Quadra 700计算机上得以实现,用于自动测定采用纸片扩散法进行的抗生素敏感性试验的抑菌圈直径。在对培养皿图像进行数字化处理后,识别并标记每个抗生素纸片。然后,利用每个纸片周围局部区域的像素生成一个轮廓图案,并对其应用决策规则。随后,将所得的抑菌圈直径估计值自动与传统断点进行比较,以便将受试菌株归类到抗生素敏感性的某一临床类别中。该程序还能够针对一些难以解读的罕见平板要求人工判读。通过将结果与对受试平板进行的独立人工测量结果相结合来测试算法的准确性。测试样本由98株菌株组成,对40种不同抗生素进行的2552次试验接受了分析。在90%的试验中,自动估计的直径与人工估计的直径之差小于4毫米。自动和人工临床分类之间的一致性达95.5%,在用葡萄球菌进行的试验中,5.6%的试验出现严重(主要和非常主要)分歧,而在用革兰氏阴性杆菌进行的试验中,只有0.3%的试验出现严重分歧。我们得出结论,径向轮廓分析算法是用于纸片扩散抗生素敏感性试验临床判读的自动系统的坚实基础。