Limits...
Automated and online characterization of adherent cell culture growth in a microfabricated bioreactor.

Jaccard N, Macown RJ, Super A, Griffin LD, Veraitch FS, Szita N - J Lab Autom (2014)

Bottom Line: While suspension culture processes benefit from decades of development of instrumented bioreactors, adherent cultures are typically performed in static, noninstrumented flasks and well-plates.A machine learning-based algorithm enabled the specific detection of one cell type within a co-culture setting, such as human embryonic stem cells against the background of fibroblast cells.In addition, the algorithm did not confuse image artifacts resulting from microfabrication, such as scratches on surfaces, or dust particles, with cellular features.

View Article: PubMed Central - PubMed

Affiliation: Department of Biochemical Engineering, University College London, London, UK Centre for Mathematics and Physics in the Life Sciences and Experimental Biology, University College London, London, UK.

Show MeSH

Related in: MedlinePlus

Automated image-processing approach. (A) Basic image features (BIFs) of the phase contrast microscopy (PCM) image are first computed. For each pixel, a local histogram of the occurrence of the different BIFs is built. These histograms are the features that are used to classify pixels as background or cells. (B) Example of a user-defined training set for the machine learning classifier. Using a conventional image-editing tool, the user indicates portions of an image that are definitely a human embryonic stem cell (hESC) colony and definitely not a colony. It is not necessary to annotate the whole image as regions can be left as not specified. (C) Schematic of the random forest classification approach. Local BIF histograms are used as inputs for decision trees. At each node, a binary test based on these features determined whether to traverse to the left or right child node next. A particular tree will classify the histogram as either cell or pixel. The majority vote of multiple trees will decide on the final class assigned to the pixel. (D) Example of processing output. (i) Binary mask after processing, showing the stem cell colony in white and the background and fibroblasts in black. (ii) Overlay of the processing results with the original PCM image.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
getmorefigures.php?uid=PMC4230958&req=5

fig2-2211068214529288: Automated image-processing approach. (A) Basic image features (BIFs) of the phase contrast microscopy (PCM) image are first computed. For each pixel, a local histogram of the occurrence of the different BIFs is built. These histograms are the features that are used to classify pixels as background or cells. (B) Example of a user-defined training set for the machine learning classifier. Using a conventional image-editing tool, the user indicates portions of an image that are definitely a human embryonic stem cell (hESC) colony and definitely not a colony. It is not necessary to annotate the whole image as regions can be left as not specified. (C) Schematic of the random forest classification approach. Local BIF histograms are used as inputs for decision trees. At each node, a binary test based on these features determined whether to traverse to the left or right child node next. A particular tree will classify the histogram as either cell or pixel. The majority vote of multiple trees will decide on the final class assigned to the pixel. (D) Example of processing output. (i) Binary mask after processing, showing the stem cell colony in white and the background and fibroblasts in black. (ii) Overlay of the processing results with the original PCM image.

Mentions: We previously developed an image-processing approach to alleviate these issues to rapidly produce accurate and reliable data, suitable for the monitoring of adherent cell culture in our microfabricated bioreactor.8 First, instead of detecting cells based on pixel intensity, we employed BIFs that can be used to classify pixels according to local symmetries.17 For example, one of the features was sensitive to dark circular objects on brighter backgrounds and thus often indicated cells’ nuclei. Similarly, the “flat” feature often indicated background regions of the image with a less marked texture. Local histograms of BIFs were constructed for each pixel of the raw PCM image (Fig. 2A). A machine learning classifier was then used to classify each pixel of an image based on its associated histogram. To do so, it was first trained by manually annotating regions of the image as either of interest or not (hESC colonies and image background/fibroblast cells, respectively, in Fig. 2B). This process is very quick as it is not required to annotate the whole image; instead, ambiguous regions can be left unannotated. This is a key advantage of the method, as image-processing methods often require extensive and tedious parameter tweaking. In this case, the algorithm can be taught how to recognize new cell types in a matter of minutes.


Automated and online characterization of adherent cell culture growth in a microfabricated bioreactor.

Jaccard N, Macown RJ, Super A, Griffin LD, Veraitch FS, Szita N - J Lab Autom (2014)

Automated image-processing approach. (A) Basic image features (BIFs) of the phase contrast microscopy (PCM) image are first computed. For each pixel, a local histogram of the occurrence of the different BIFs is built. These histograms are the features that are used to classify pixels as background or cells. (B) Example of a user-defined training set for the machine learning classifier. Using a conventional image-editing tool, the user indicates portions of an image that are definitely a human embryonic stem cell (hESC) colony and definitely not a colony. It is not necessary to annotate the whole image as regions can be left as not specified. (C) Schematic of the random forest classification approach. Local BIF histograms are used as inputs for decision trees. At each node, a binary test based on these features determined whether to traverse to the left or right child node next. A particular tree will classify the histogram as either cell or pixel. The majority vote of multiple trees will decide on the final class assigned to the pixel. (D) Example of processing output. (i) Binary mask after processing, showing the stem cell colony in white and the background and fibroblasts in black. (ii) Overlay of the processing results with the original PCM image.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2 - License 3
Show All Figures
getmorefigures.php?uid=PMC4230958&req=5

fig2-2211068214529288: Automated image-processing approach. (A) Basic image features (BIFs) of the phase contrast microscopy (PCM) image are first computed. For each pixel, a local histogram of the occurrence of the different BIFs is built. These histograms are the features that are used to classify pixels as background or cells. (B) Example of a user-defined training set for the machine learning classifier. Using a conventional image-editing tool, the user indicates portions of an image that are definitely a human embryonic stem cell (hESC) colony and definitely not a colony. It is not necessary to annotate the whole image as regions can be left as not specified. (C) Schematic of the random forest classification approach. Local BIF histograms are used as inputs for decision trees. At each node, a binary test based on these features determined whether to traverse to the left or right child node next. A particular tree will classify the histogram as either cell or pixel. The majority vote of multiple trees will decide on the final class assigned to the pixel. (D) Example of processing output. (i) Binary mask after processing, showing the stem cell colony in white and the background and fibroblasts in black. (ii) Overlay of the processing results with the original PCM image.
Mentions: We previously developed an image-processing approach to alleviate these issues to rapidly produce accurate and reliable data, suitable for the monitoring of adherent cell culture in our microfabricated bioreactor.8 First, instead of detecting cells based on pixel intensity, we employed BIFs that can be used to classify pixels according to local symmetries.17 For example, one of the features was sensitive to dark circular objects on brighter backgrounds and thus often indicated cells’ nuclei. Similarly, the “flat” feature often indicated background regions of the image with a less marked texture. Local histograms of BIFs were constructed for each pixel of the raw PCM image (Fig. 2A). A machine learning classifier was then used to classify each pixel of an image based on its associated histogram. To do so, it was first trained by manually annotating regions of the image as either of interest or not (hESC colonies and image background/fibroblast cells, respectively, in Fig. 2B). This process is very quick as it is not required to annotate the whole image; instead, ambiguous regions can be left unannotated. This is a key advantage of the method, as image-processing methods often require extensive and tedious parameter tweaking. In this case, the algorithm can be taught how to recognize new cell types in a matter of minutes.

Bottom Line: While suspension culture processes benefit from decades of development of instrumented bioreactors, adherent cultures are typically performed in static, noninstrumented flasks and well-plates.A machine learning-based algorithm enabled the specific detection of one cell type within a co-culture setting, such as human embryonic stem cells against the background of fibroblast cells.In addition, the algorithm did not confuse image artifacts resulting from microfabrication, such as scratches on surfaces, or dust particles, with cellular features.

View Article: PubMed Central - PubMed

Affiliation: Department of Biochemical Engineering, University College London, London, UK Centre for Mathematics and Physics in the Life Sciences and Experimental Biology, University College London, London, UK.

Show MeSH
Related in: MedlinePlus