Figure 1.
Temporal variation of the morphological characteristics of MTB cords in MODS culture observed at 100X total magnification.
(A) Day 3 of culture; (B) Day 10 of culture; (C) Day 15 of culture. Cords observed at day 10 are highly specific for MTB.
Figure 2.
Digital photos of MODS cultures.
Corresponding to: (A, B) Mycobacterium tuberculosis; (C) Objects from sediment and detritus of a mycobacterium negative sample; (D) Mycobacterium kansasii; (E) Mycobacterium avium; (F) Mycobacterium chelonae.
Figure 3.
Flowchart of the image processing algorithm.
The original photo (in yellow), is processed by the ‘image processing algorithm’ (in green). The numbers inside the green boxes correspond to the steps described in the manuscript. Five images are obtained: Gray scale image, SKL image, Borders image, Cleaned image and finally the Tri-colored image, which is built using the Cleaned image and Borders image.
Figure 4.
Steps of image processing of a MODS-culture photo.
(A) Original photo. (B) Conversion to gray scale, obtaining the ‘Gray scale image’. (C) Global binarization, transforming all the pixels in black (0) or white (1). (D) Black-white inversion. (E) Border smoothing, to reduce the ‘noise’. The yellow rectangle in figures D (before) and E (after) shows the changes produced by this process. (F) Exclusion of boundary objects, deleting the objects in the border of the photo. The green rectangle in figures E (before) and F (after) shows the changes produced by this process. (G) Holes Filtering, removing black objects inside white objects. The blue rectangle in figures F (before) and G (after) shows the changes produced by this process. (H) Area filtering, dropping the outsider values in the distribution of area values, obtaining the ‘Cleaned image’. (I) Skeletonization, obtaining the ‘SKL values’. (J) Identification of object borders, obtaining the ‘Borders image’. (K) Image recoloring: The black pixels in figure H are converted to gray, and this picture is superposed by the figure J, giving the ‘Tri-colored image’. Figures (B), (I) and (K) are used for the features extraction process.
Figure 5.
Morphological characteristics of a Mycobacterium tuberculosis cord from a TB-positive MODS culture.
The process performed to obtain the main features of a M. tuberculosis cord includes the following steps: (A) Skeleton of the object, (in purple an example of end-points, in red branch-points, and in green inner-points); (B) Identification of the trunk (gray) and its branches (bold black); (C) Extension of the skeleton. The extension of the trunk (in red) from each end-point of the trunk to its nearest tip of the object (enclosed in purple), and the border of the object (bold black); (D) The extended trunk of the object divided by equally-spaced segments (in red, on the gray skeleton); (E) Transversal division of the object by segments (in cyan) perpendicular to and equally spaced along the extended trunk; (F) The “linearization” of the extended trunk (red line) and waves recognition (one wave is enclosed in green as example).
Figure 6.
Flowchart of the features extraction algorithm.
The SKL image, the Tri-colored image and the Gray scale image are used as input in different steps of the process (in orange) to extract the features. From this process 53 features are obtained for the statistical analysis.
Table 1.
Dictionary of variables.
Table 2.
Univariate analysis of the main features relevant to the prediction of TB positive objects.
Figure 7.
Sensitivity and specificity of the best model to diagnose Mycobacterium tuberculosis from a MODS culture digital image.
(A) Sensitivity and specificity for different probability cutoffs for the best object-model to classify M. tuberculosis cords; (B) Sensitivity and specificity for different probability cutoffs for the best photo-model to classify M. tuberculosis positive MODS culture images.
Table 3.
Relevance of features from the best photo model in a univariate and multiple variable logistic regression.
Table 4.
Relevance of features from the best two variants object models, in a univariate and multiple variable logistic regression.