Method for detecting colonies of microorganisms in a sample arranged in a culture medium

Abstract

The invention relates to a method for detecting colonies of microorganisms in a biological sample arranged in a solid culture medium, the method being implemented by a detection system comprising at least one incubator, an analysis unit, and at least one automated image capture system, the method comprising a first phase of detecting and counting objects, in the field of view of the image capture system, exhibiting growth, and a second phase applied only to those samples deemed to be negative in the first phase of the method, the second phase being based on anomaly detection applied to at least the last image acquired in the incubation sequence.

Description

[0001] Method for detecting colonies of microorganisms in a sample placed in a culture medium

[0002] Technical field of the invention

[0003] The invention finds its application in particular in the field of microbiological control of sterile products and their production environment. The invention relates to the field of contamination detection systems on Petri dishes or other culture media, preferably solid, to be incubated. More particularly, the invention relates to the field of colony forming unit (CFU) detection systems on Petri dishes and based on sequential and automated reading of said Petri dishes throughout the incubation period.

[0004] Technological background of the invention

[0005] Environmental microbiological monitoring of production areas is regularly carried out in the pharmaceutical field, particularly for grade A and B controlled aseptic production areas. Traditionally, sedimentation dishes are used for air analysis or so-called "contact" dishes for surface analysis, the incubation time of which varies between 5 and 7 days and the reading of the result (contamination or not) is carried out after incubation by an operator. This type of monitoring of aseptic areas poses particular constraints insofar as: i. several hundred Petri dishes containing a solid culture medium can be recovered each day depending on the defined control plan and regulatory requirements, ii. most of the samples taken in these areas are negative (between 95 and 99.9%), iii.more than the exact number of colonies growing on a solid culture medium, it is the qualitative status of the sample (negative for CFU=0 and positive for CFU>0), which is monitored, as recommended in the regulatory texts ("Contamination Recovery Rate defined in the USP <1116> ).

[0006] In the event of contamination, an investigation is carried out to identify the contaminating microorganism(s) in order to know their species and / or strain and thus trace the causes of the contamination.

[0007] This traditional method is relatively long, tedious, difficult to trace and dependent on the operator's skills (training, visual ability, etc.). Even if it is considered the reference, the performance of this method is not perfect and there remains an estimated false negative rate of around 2%.

[0008] To make this method more efficient and traceable, without degrading its performance, automated reading systems based on digital image capture and associated processing have been developed.

[0009] On the one hand, there are conventional colony reading systems, called "endpoint" where human reading is replaced by recording and automated processing of images taken on the Petri dish at the end of incubation. These systems are well suited for enumeration applications, when the microbial load (quantity of microorganism in the sample) is generally significant (a few dozen or even hundreds of colony-forming units) and the aim is not to perform a qualitative presence / absence test. However, in the case of qualitative contamination detection applications, the performance is generally not as good as that of the traditional method in terms of sensitivity (ability to correctly detect a colony when it is present) and specificity (ability not to trigger a false alarm).

[0010] There are indeed many artifacts on the Petri dish (dust, marks, defects on the agar surface, etc.) that are wrongly recognized as colonies, regardless of the quality of the optical capture system, the image processing or the algorithm used. There is a trade-off between false positives and false negatives inherent in the acquired basic information not compatible with the requirements of the qualitative contamination testing application, as illustrated in Figure 1.

[0011] To overcome the limitations of automatic "end-point" reading systems and to consider equivalence with the performance of the traditional method, particularly for environmental microbiological control of aseptic production areas, a new type of system has been developed, called "continuous reading", integrating an incubator, and based on automated reading, by taking images throughout the incubation and processing the sequences thus obtained.

[0012] These systems read from closed Petri dishes to avoid cross-contamination inside the incubator containing multiple Petri dishes and generally integrate multiple contrast sources (transmission, diffuse reflection, grazing illumination, etc.) in order to capture the maximum amount of information and compensate for possible reading biases due to the presence of the lid (which can be removed in the traditional method of reading at the end of incubation). In these systems, successive images of the Petri dish are recorded during incubation, generally at regular intervals (e.g., one image every hour) and their analysis is carried out as the acquisition progresses, with image processing operations and a generic protocol as described in Figure 2.

[0013] These systems integrate an image processing chain with some critical steps, such as the normalization of intensities (in order to eliminate variability due to lighting drifts or changes in the growth medium), or the registration of an image with respect to the previous one, allowing the matching of detected objects and the calculation of the growth criterion.

[0014] Important parameters of the algorithm that allow adjusting the slider between sensitivity and specificity include:

[0015] -the thresholding criterion for segmenting and extracting an object above the background,

[0016] -the variation in size of the detected object allowing growth to be declared between two images,

[0017] -the number of successive growths of an object to declare it the colony-forming unit.

[0018] These continuous reading systems allow significant performance improvements compared to endpoint reading systems because the growth criterion is universally present for colonies and absent for artifacts, whose size generally does not vary, or rather randomly.

[0019] However, despite the clear improvement brought by continuous reading systems, there remain cases, marginal admittedly, of false positives and false negatives. For example, a colony that starts its growth in a masked area (edge ​​of Petri dish, shadow or close to a fixed particle) will not be detected by such an algorithm because it is quickly annotated as an artifact (first images after the appearance without the minimum number of successive growths). Similarly, a shadow whose size increases over a few images will be annotated as a colony. There is obviously a possible adjustment in the algorithm parameters (segmentation threshold, size variation threshold and number of successive growths) which allows to detect a colony with an atypical growth behavior, but at the cost of an increase in the false positive rate. There therefore remains a residual rate of false positives and false negatives inherent to the growth detection method.

[0020] These false negatives and false positives remain a major problem for environmental monitoring applications in aseptic areas: currently, on this type of continuous reading system, the false positive rate must be maintained around 5 to 10% in order to guarantee a false negative rate equivalent to the traditional method (around 2%), which remains problematic to justify the economic gain of automating the traditional method.

[0021] Subject of the invention

[0022] The invention aims to remedy all or part of the aforementioned drawbacks and in particular to further reduce the number of false negatives and / or false positives in the context of continuous reading systems for Petri dishes.

[0023] To this end, the subject of the invention is a method for detecting colonies of microorganisms in a sample deposited in a solid culture medium, said method being implemented by a detection system comprising at least one incubator, an analysis unit, and at least one automated image capture system, said method comprising a first phase of detecting and counting objects, in the field of view of the image capture system, exhibiting growth, the first phase comprising at least the steps according to which:

[0024] •Acquisition of a plurality of images of the culture medium in the field of view of the image capture system, during the incubation of said sample in the incubator, by the image capture system,

[0025] •Detection of objects in images acquired by segmentation with a thresholding method;

[0026] •Extraction on said detected objects of a parameter of growth of size of the object between successive images,

[0027] •Determination of the presence or absence of microorganism(s) by detection of growth of said at least one object by the analysis unit, (i) if growth of the object is detected, then the analysis unit considers that there is presence of at least one colony of microorganisms in the sample, the sample is said to be positive; (ii) if no growth of the object is detected, then the analysis unit considers that there is absence of colony of microorganisms in the sample, the sample is said to be negative,

[0028] • Counting of said objects exhibiting growth by the analysis unit, characterized in that the method further comprises a second phase applied only to the samples considered negative in the first phase of the method, said second phase being based on an anomaly detection applied to at least the last image acquired during the incubation sequence, and comprising the following steps:

[0029] •Detection of objects in the acquired image by segmentation with a thresholding method; •Extraction on said detected objects of position and / or size and / or morphology and / or intensity parameters,

[0030] •Application of a discrimination model estimated from a learning database contained in the analysis unit for example and based on a score calculated from all the parameters extracted in the extraction step in order to separate anomalies from artifacts.

[0031] This process has several advantages detailed below.

[0032] The method according to the invention makes it possible in particular to gain in specificity (capacity not to trigger false alarms) in the first phase even if it means reducing the sensitivity (capacity to detect colonies), the sensitivity being guaranteed thanks to the second phase, which ultimately makes it possible to reduce the rate of false-positive results in addition to the classic objective of reducing the rate of false-negative results.

[0033] The method according to the invention aims only to identify anomalies which are associated with the presence of one or more colonies: the classification criteria for these anomalies are different from those which are conventionally used for automated reading of Petri dishes at the end point and include in particular the position of the colony. Indeed, an object detected by the second phase, at a distance from the edges of the dish (center of the dish) and not detected by the first phase of the algorithm has a greater probability of being an artifact.

[0034] The second phase applies primarily to the last image acquired from the sequence at the end of incubation, the colonies present at this stage are generally clearly visible with sizes exceeding 500pm, and therefore the anomalies are clearly identifiable, whereas in the first phase the detection takes place as they appear, the objects to be detected are much smaller, of the order of a few tens of microns,

[0035] Furthermore, the second phase also has a significant impact on the development of the model of the said first phase. Indeed, in the current state of the art of continuous reading detection / counting systems, the model is optimized with an initial constraint in terms of false negatives (the greatest risk being to let contamination pass). The target rate is determined around 2% according to the error estimates of the traditional method. From this target rate, the algorithm is optimized on a basis of image sequences containing all possible cases, and in particular the so-called "difficult" cases of colonies growing at the edges of dishes, with low contrast and / or slow growth.The algorithm criteria (threshold for segmenting an object and extracting it from the background, size variation rate between 2 images from which a growth is declared and number of successive growths to annotate an object as a colony-forming unit) are adapted to achieve this rate on the complete basis of all possible cases. This “optimal” model on the entire basis or can be “sub-optimal” for the basis composed only of simple cases. To the extent that, in the principle of the invention, the second phase is applied only to samples considered negative at the end of the first phase, the optimization criterion of the first phase can be modified, in particular by modifying the calculation basis on which the false negative rate is estimated: we remove the cases of detection a priori difficult on the growth criterion, knowing that they will be recovered by the second phase specifically dedicated.These cases are removed from the base of the first phase model on the dual condition that:

[0036] 1. removing them improves performance (reduction in the false positive rate and the false negative rate) on the limited basis of “simple” cases;

[0037] 2. the presence of the colony generates an anomaly clearly visible on the last image, even if the colony is not perfectly identifiable.

[0038] In the present application, the term "object" designates an element detected on an image via the thresholding / segmentation algorithm of the analysis unit. The two phases of the detection method are both based on a first step of object detection. However, it is not necessarily the same objects that are detected because (i) on the one hand the algorithm is not necessarily the same (ii) on the other hand the contrast mode used (or contrast fusion) is not necessarily the same and (iii) finally and above all, the moment when the object is detected is not the same: at the appearance during incubation (small objects) and on the final point image for the second phase of the algorithm (larger objects a priori).

[0039] In the present invention, the term "anomaly" refers to an object detected during the second phase of the algorithm and which is classified as a colony-forming unit. The detected object may not resemble a colony and in fact correspond to a "merger" between a colony and a box edge, or even several merged colonies, hence the term "anomaly".

[0040] In the present invention, the term "artifact" refers to an object detected but not considered as a colony-forming unit by the algorithm (rightly or wrongly). Similarly, there are artifacts both in the first phase but possibly in the second phase with objects that will be detected on the final point image, but ultimately not classified as colony-forming units with the discrimination model.

[0041] According to a characteristic of the invention, the second phase may comprise one or more additional pre-processing steps such as the fusion of different contrast modalities acquired at the end point, conversion into gray levels (8 bits or more), normalization of the intensities of the image to compensate for possible variations in lighting, filtering by convolution, definition of a region of interest.

[0042] According to a characteristic of the invention, the morphological parameter may be the convexity of the detected object and / or the circularity of the detected object. Advantageously, the morphological classification parameters are less strict than those of position insofar as one can have an object extracted on the image consisting of the fusion between a colony (rather circular and convex object) and a mark, as illustrated in figure 4. The sensitivity and specificity objectives of this anomaly detection layer are therefore very distinct from those of a conventional application of colony detection on an image.For example, if we consider a normalized morphological parameter such as the convexity of the object with a range between 0 and 1, (1 for a perfect disc and 0 for example a half-moon), and a position criterion also normalized such as the relative distance from the center (0 if the object is in the center and 1 if it is on the extreme edge of the culture medium); then if a model with acceptance ranges is used, we will have, for the second phase for example:.

[0043] Anomaly detection if the position criterion is between 0.8 and 1 (therefore very eccentric object) and convexity criterion between 0.3 and 1 (wider acceptance range because it is possible to have two merged colonies which will not form a disc but a slightly concave shape) for example, and Artifact detection for anything that is detected and considered not to be an anomaly

[0044] According to a characteristic of the invention, the size parameter can be the diameter and / or the perimeter and / or the area of ​​the detected object.

[0045] According to a characteristic of the invention, the intensity parameter can be the average gray levels of the detected object and / or the variation of the intra-object gray level.

[0046] According to a characteristic of the invention, the score calculated from all the extracted parameters, allowing the classification of the object, can be calculated from one of the methods: regression, decision tree, linear discriminant analysis, Support Vector Machine, Neural Networks. Advantageously, from the different criteria calculated on the extracted objects, the discrimination model is therefore estimated from a restricted and specific learning base containing only the colonies not detected by the first phase and all possible artifacts, which makes it possible to recover the missed colonies without increasing the false alarms.

[0047] According to a characteristic of the invention, the second phase may further comprise a step of merging the different contrast modalities, and a step of conversion into gray levels, for example on 8 bits in 256 levels, these steps being carried out before the step of segmenting the objects.

[0048] According to a characteristic of the invention, the invention can be applied to fluorescence image captures.

[0049] According to a characteristic of the invention, when the sample is considered positive by the analysis unit, whatever the phase, an alert is transmitted to the user.

[0050] According to a characteristic of the invention, the parameterization of the first phase is optimized on a database of nominal cases by excluding problematic cases, in particular colonies at the edges of the culture medium, etc.).

[0051] In this application, "a case" means a colony present in the image sequence or an artifact present in the image sequence.

[0052] The database is built from a plurality of inoculated or uninoculated culture media. Depending on the average inoculation level (e.g., five Colony Forming Units (CFU)), the database can include approximately one thousand cases (colonies or artifacts), as there are multiple cases per culture medium.

[0053] Among these cases in the database (colonies or artifacts) we distinguish two categories: (i) nominal cases and (ii) marginal cases. A marginal case can also be called a "problematic case". Only marginal cases are specified, and all cases that are not marginal are by definition nominal cases.

[0054] These marginal cases correspond to cases of colonies or artifacts with characteristics such as incorrect prediction and false positives (artifacts wrongly predicted by the algorithm as CFUs, or conversely true CFUs missed by the algorithm considered as artifacts).

[0055] According to the invention, the marginal cases correspond to:

[0056] - a colony that starts growing in a hidden area (edge ​​of Petri dish, shadow or near a fixed particle)

[0057] - a shadow that increases in size over a few images and therefore risks being wrongly detected as a UFC

[0058] Nominal cases are all other cases, namely colonies that do not start their growth in a masked area or near a fixed particle and artifacts whose appearance does not vary from one image to another.

[0059] According to a characteristic of the invention, the second phase further comprises a step of pre-processing the last acquired image involving another image of the sequence, for example the first acquired image. Brief description of the figures

[0060] The invention will be better understood from the following description, which relates to embodiments according to the present invention, given as non-limiting examples and explained with reference to the attached schematic figures. The attached schematic figures are listed below:

[0061] Figure 1 is a view of the different current systems (a), (b) and (c) for reading Petri dishes.

[0062] Figure 2 is a view of the general processing procedure of so-called continuous automatic reading systems.

[0063] Figure 3 is a view of the general method of the invention with two-phase detection.

[0064] Figure 4 is a view of the principle of the second phase of anomaly detection.

[0065] Figure 5 is a diagram of the method according to the invention.

[0066] Figure 6 is a view explaining the impact of the presence of the second phase in the optimization criteria of the first phase of the algorithm.

[0067] Figure 7 is an illustration of a first image processing method for object detection according to the invention.

[0068] Figure 8 is an illustration of a second image processing method for object detection according to the invention.

[0069] Figure 9 illustrates a detection system according to the invention.

[0070] Detailed description

[0071] Figure 1 shows the state of the art of colony detection and counting systems. Schematically, a Petri dish (3) with an artifact (2) and a colony (1) at the edge of the Petri dish (3) are illustrated, along with a graph showing the performance of current methods (a), (b), and (c) with a False Negative rate of around 2%.

[0072] In (a) is illustrated the traditional method consisting of an examination by a qualified operator, at the end of incubation, without a cover. The disadvantages of such a method lie in the fact that the examination is variable geometry and relies on human interpretation. This traditional method has good performance, however this method is long, tedious and not very traceable.

[0073] In (b) and (c) are illustrated automated methods aiming to maintain the performance of the traditional method (a). In (b) is illustrated an automated end-point reading method implemented by a camera-based optical system, carried out at the end of incubation, with or without a cover. The disadvantages of such a method lie in the fact that the examination is simple with image processing: it is therefore recurrent that the artifact is detected as a colony due to the lack of complex geometry and a growth examination. When examining the graph relating to method (b), it is noted that it is not possible to reach a compromise (hatched area referenced 4 on the graph) between sensitivity (z) and specificity (w) compatible with the application, the AA axis representing the decision threshold.

[0074] In (c) is illustrated an automated reading method in kinetics implemented by an optical system based on a camera, carried out throughout the incubation, with cover (5). The disadvantages of such a method lie in particular in the fact that the examination is simple with image sequence processing: it is therefore recurrent that a colony (1) at the edge of the box is not detected because its growth begins in a peripheral zone (edge ​​of the box) which is masked and therefore which masks the appearance of the colony. When we examine the graph relating to method (c), we see that the compromise between sensitivity (z) and specificity (w) is much better than in (b) but that there are still cases to be improved: growth in the peripheral zone, start of masked growth, etc. (hatched zone referenced 4 on the graph).

[0075] Figure 2 represents the general procedure of the current detection and counting methods (c). Figure 2 illustrates a plurality of images of a Petri dish (3) acquired throughout the incubation (represented by the arrow above). This plurality of acquired images will be recorded (step 1) then an object detection will be carried out (step 2), a dimension extraction will also be done (step 3) and a similar object association will be carried out (step 4) following these four steps, it is possible to carry out a growth analysis (step 5) represented in Figure 3 by the graph.

[0076] In Figure 3, the y-axis represents the size of the detected object and the x-axis represents time. Returning to Figure 2, according to the growth analysis, the element referenced (1) is identified as a colony while the element referenced (2) is identified as an artifact, given that the latter does not show any change in size over time.

[0077] Figure 4 illustrates the principle of anomaly detection associated with a CFU.

[0078] The first image (a) is a photo of a Petri dish (3) in end point. A colony that has grown at the edge of the dish is circled (3). From this image, the analysis unit performs processing and in particular segmentation which allows the objects to be highlighted, as can be seen in (b). The surrounded object (2) is considered in its entirety and includes the colony that has grown at the edge of the dish because its outline is confused with an object in the dish. In a classic detection method, the analysis unit's algorithm that allows the determination of the presence or absence of a CFU considers the colony (1) as an artifact (2) because the latter is included in the shape of the artifact and the resulting shape does not correspond to the shape criteria of a colony nor to the growth criteria of a colony.According to the invention, this same object is detected as an anomaly associated with a CFU even with a morphology close to that of an artifact thanks to the steps of the method according to the invention.

[0079] Figure 5 represents the detection method according to the invention. Said detection method according to the invention (200) is implemented by a detection system (300) represented in Figure 9 and comprising at least one incubator (301), an analysis unit (302), and at least one automated image capture system (303).

[0080] As can be seen in Figure 5, the detection method comprises a first phase (PI) of detecting and counting objects exhibiting growth, in the field of view of the image capture system (303), the first phase comprising at least the steps according to which:

[0081] (P101) - Acquisition of a plurality of images of at least one object in the field of view of the image capture system (303), during the incubation of said sample in the incubator, by the image capture system,

[0082] (P102) - Determination of the presence or absence of microorganism(s) by detecting growth of said at least one object by the analysis unit (302), (i) if growth of the object is detected, then the analysis unit considers that there is presence of at least one colony of microorganisms in the sample, the sample is said to be positive; (ii) if no growth of the object is detected, then the analysis unit considers that there is absence of colony of microorganisms in the sample, the sample is said to be negative,

[0083] (P103) - Counting of said objects showing growth by the analysis unit, (P300) - Alerting the user of the number / presence of samples considered positive.

[0084] Furthermore, as illustrated in Figure 5, the method according to the invention further comprises a second phase (P2) applied only to the samples considered negative (ii) in the first phase (PI) of the method. The second phase (P2) is based on an anomaly detection applied to at least the last image acquired during the incubation sequence, and comprising the following steps: (P201) Detection of objects in the acquired image by segmentation with a thresholding method;

[0085] (P202) Extraction on said detected objects of position and / or size and / or morphology and / or intensity parameters,

[0086] (P203) Application of a discrimination model estimated from a training database contained in the analysis unit and based on a score calculated from all the parameters extracted in the extraction step in order to separate anomalies from artifacts.

[0087] (P204) the score is compared to a threshold value which makes it possible to determine whether there is an anomaly, i.e. a colony-forming unit which should have been detected (case (i)) or whether it is an artefact (case (ii)). In case (i), an alert is activated to warn the user of the presence of a colony-forming unit and therefore of the presence of microorganisms and in case (ii), the sample is considered negative: absence of microorganism.

[0088] Figure 6 shows the impact of the presence of a second anomaly detection phase after a first phase.

[0089] In (a) of Figure 6, a classical approach is shown, with continuous reading, a sub-optimal performance is achieved. Here, to guarantee the good segmentation of the “difficult” colony at the edge of the box (1), the threshold must be adapted to the detriment of the detection of a “normal” colony (2) appearing in a simple area.

[0090] In (b) and (c) of Figure 6 is shown a two-phase approach according to the invention. Unlike the classical approach illustrated in (a), the model of the first phase is clearly optimized for “perfect” detection in simple conditions (2), and the “missed” colony (1) because it is close to a marked area will be excluded from the learning base (see (b)). The “missed” colony (1) in the first phase (b) will be recovered in the second phase (c) because it is clearly visible at the last image of the acquired image sequence.

[0091] Figure 7 illustrates a first image processing method for object detection comprising: a first image processing, namely segmentation (b) on the acquired image (a) then a second processing, namely discrimination (c).

[0092] Figure 8 illustrates a second image processing method for object detection comprising: the first image of the acquired sequence (d) and the last image of the acquired sequence (f), the processing making it possible to differentiate between the two images (g), followed by discrimination (h). Of course, the invention is not limited to the embodiments described and represented in the appended figures. Modifications remain possible, in particular from the point of view of the constitution of the various elements or by substitution of technical equivalents, without departing from the scope of protection of the invention.

Claims

CLAIMS 1. Method for detecting colonies of microorganisms in a sample deposited in a solid culture medium, said method being implemented by a detection system comprising at least one incubator, an analysis unit, and at least one automated image capture system, said method comprising a first phase (PI) of detecting and counting objects, in the field of view of the image capture system, exhibiting growth, the first phase (PI) comprising at least the steps according to which: (P101) Acquisition of a plurality of images of the culture medium in the field of view of the image capture system, during the incubation of said sample in the incubator, by the image capture system, (P102) Object detection in images acquired by segmentation with a thresholding method; Extraction on said detected objects of a parameter of growth of size of the object between successive images, Determination of the presence or absence of microorganism(s) by detecting growth of said at least one object by the analysis unit, (i) if growth of the object is detected, then the analysis unit considers that there is presence of at least one colony of microorganisms in the sample, the sample is said to be positive; (ii) if no growth of the object is detected, then the analysis unit considers that there is absence of colony of microorganisms in the sample, the sample is said to be negative, (P103) Counting said objects exhibiting growth by the analysis unit, characterized in that the method further comprises a second phase (P2) applied only to the samples considered negative in the first phase of the method, said second phase (P2) being based on an anomaly detection applied to at least the last image acquired during the incubation sequence, and comprising the following steps: (P201) Detection of objects in the acquired image by segmentation with a thresholding method; (P202) Extraction on said detected objects of position and / or size and / or morphology and / or intensity parameters, (P203) Application of a discrimination model estimated from a learning database contained in the analysis unit for example and based on a score calculated from all the parameters extracted in the extraction step in order to separate anomalies from artifacts.

2. Detection method according to claim 1, in which the second phase may comprise one or more additional pre-processing steps such as the fusion of different contrast modalities, conversion to gray levels, convolution filtering, definition of a region of interest.

3. Detection method according to any one of claims 1 or 2, wherein the morphological parameter may be the convexity of the detected object and / or the circularity of the detected object.

4. Detection method according to any one of claims 1 to 3, in which the morphological classification parameters are less strict than those of position.

5. Detection method according to any one of claims 1 to 4, wherein the size parameter may be the diameter and / or the perimeter and / or the area of the detected object.

6. Detection method according to any one of claims 1 to 5, wherein the intensity parameter may be the average gray levels of the detected object and / or the variation of the intra-object gray level.

7. Detection method according to any one of claims 1 to 6, in which the score calculated from all the extracted parameters, allowing the classification of the object, can be calculated from one of the methods: regression, decision tree, linear discriminant analysis, Support Vector Machine, Neural Networks.

8. Detection method according to any one of claims 1 to 7, in which from the different criteria calculated on the extracted objects, the discrimination model is therefore estimated from a restricted and specific learning base containing only the colonies not detected by the first phase and all possible artifacts.

9. Detection method according to any one of claims 1 to 8, in which the second phase (P2) further comprises a step of merging the different contrast modalities, and a step of conversion into gray levels, for example on 8 bits in 256 levels, these steps being carried out before the step of segmenting the objects.

10. Detection method according to any one of the preceding claims, in which the parameterization of the first phase is optimized on a database of nominal cases by excluding problematic cases.

11. Detection method according to any one of the preceding claims, in which the second phase further comprises a step of pre-processing the last acquired image involving another image of the sequence, for example the first acquired image.

Images