Job: Micrograph Junk Detector (BETA)
Last updated
Last updated
Automatically detect contaminants in a micrograph and remove particle picks near those contaminants.
The Micrograph Junk Detector is a pre-trained neural network model that analyzes micrographs and identifies contaminants in several distinct classes. Broadly defined, the junk detector tries to classify each region of the micrograph by whether it is:
Good ice (where particles may be found)
Carbon or gold support (i.e., the edge of a hole)
An intrinsic ice defect (e.g., crystalline ice)
An extrinsic contaminant or ice defect (e.g., a large crystal of transfer ice or a hydrocarbon contaminant)
For each micrograph, the Micrograph Junk Detector produces a junk annotation mask, labelling each point in the micrograph with one of the above classes. From the junk annotation, statistics such as the fraction of junk area in the micrograph image and the ratio of different types of junk are computed. The junk annotation and junk statistics are both stored with the micrographs output from the Micrograph Junk Detector and can be used to filter and sort them downstream. Furthermore, particle picks which are on or near the labelled junk contaminants can be discarded.
Micrographs connected to this slot will be analyzed. If the input micrographs already have junk annotations, they will not be re-processed. Instead, their existing junk annotations will be used to filter particles (if particles are connected).
Particles connected to the Micrograph Junk Detector which will be filtered by their proximity to junk.
This parameter has no effect if particles are not connected.
A particle is rejected if its center is this close or closer to a pixel on the micrograph which has been labeled as any type of junk. Distances for each of the three types of junk can be adjusted using the advanced type-specific parameters of the Micrograph Junk Detector.
Otherwise, the micrographs are unchanged from the input.
Particles which were not near or inside junk. These particles are generally expected to contain fewer off-target picks.
Particles which were within any of the minimum junk distances. Note that particles are not grouped by which type of junk they were close to.
The first plot produced by the Micrograph Junk Detector counts the number of micrographs with any pixels labeled as each of the three types of junk (bar chart, left). For example:
If 100% of the micrographs have 10% of their area covered by extrinsic ice, the Ext. Ice bar will be at 100%.
If 10% of the micrographs are entirely full of extrinsic ice, the Ext. Ice bar will be at 10%.
The next plot produced by the Micrograph Junk Detector displays the total area of each junk type for all of the micrographs analyzed (donut chart, right). Both of the examples given above (100% of the micrographs having 10% extrinsic ice, or 10% having 100% contamination) would result in the Extrinsic Ice Defect region occupying 10% of the donut.
These two example plots come from the same Micrograph Junk Detector job. Note that 3 of the micrographs have some intrinsic ice defects, but the detected region is so small that it registers as 0.1% of the micrograph area plot.
Finally, the Micrograph Junk Detector generates images of the first 20 micrographs processed. The micrographs are annotated with colors corresponding to the three junk types (or no color for good ice).
If particles are connected, a second image is generated showing particle positions and whether they were rejected (red circle with a transparent fill) or accepted (white circle with no fill).
Micrographs with junk annotations. These annotations can be viewed and used to filter whole micrographs in . Additionally, the annotations are used to accept or reject particles in subsequent Micrograph Junk Detector jobs.
The Micrograph Junk Detector is most often used between particle picking (like or ) and to avoid extracting junk picks. It can also be used to reject particle images which have already been extracted.
Additionally, now includes a % Junk filter to eliminate micrographs which are mostly or entirely junk before particle picking. Manually Curate Exposures can also be used to visually investigate individual micrograph’s junk areas.
