Changes between Version 8 and Version 9 of barSoftwareErrorCorrection


Ignore:
Timestamp:
10/25/11 16:30:58 (13 years ago)
Author:
kubel
Comment:

--

Legend:

Unmodified
Added
Removed
Modified

  • barSoftwareErrorCorrection

    v8 v9  
    1 == Typical problems in vector reconstruction and their solutions == 
     1== Preparation of CAF slide from vector graphics == 
    22 
     3=== ''Contour slide''  === 
     4 
     5In our work-flow we introduced a stage of ''contour slides'' - vector graphic in SVG format (for detailed description see Majka et al 2011) that simplifies necessary corrections, manual preparation of structures' outlines from histology results and further improvements of slide details.  
     6 
     7Left: Coronal slice of a rat brain around 3 mm behind Bregma stained for Cytochrome Oxidase. Histological verification of the electrodes' location after chronic LFP recording from the thalamic somatosensory nuclei. 
     8 
     9Middle: Initial stage of ''countour slide'' creation. Main structures outlines drawn over the histology slice (line thickness increased for better visibility) and structure labels placed over well recognized regions. 
     10 
     11Right: Histological slide visible through translucent final CAF slide. 
     12 
     13 
     14[[Image(006__final_comparison.gif)]] 
     15 
     16---- 
     17== Typical problems and their solutions == 
    318Most commonly available 2D atlases were not designed to be used as 
    419sources for 3D models generation.  Precision of contour definitions 
    … …  
    1126reconstruction, indeed for any systematic computer processing. 
    1227 
    13  
    14 [[Image(001__unlabelled_annotated.gif)]] 
    15 [[Image(002__gapfill_annotated.gif)]] 
    16 [[Image(003__invalid_labels_annotated.gif)]] 
    17 [[Image(006__final_comparison.gif)]] 
    18 [[Image(007__final_electrode.gif)]] 
    19  
    20 [[Image(0041_new_edges_animated.gif)]] 
     28Manually prepared ''contour slides'' - including our example  
     29presented here - also can contain errors requiring corrections.  
    2130 
    2231=== Non-interactive error correction === 
    … …  
    2837name of the problematic label and its location. These data help the 
    2938user to solve the problem manually by removing ambiguities from 
    30 contour slides. Below we discuss common inconveniences and our 
     39contour slides. Below we discuss and illustrate common inconveniences and our 
    3140strategies for fixing them. 
    3241 
    33 === Leaking structures === 
    3442 
    35 The most common defect is gaps in structure outlines. It frequently 
    36 happens when two contour lines, which should touch are drawn so that 
    37 they leave a little space in between. Such an arrangement may not be 
    38 visible in a printed atlas but it greatly disturbs the tracing process 
    39 where every pixel may influence the results. In this case, the 
    40 structure being traced overtakes the space of its neighbour through 
    41 the broken contour which we call ''leaking of the structure'' '''(see 
    42 the spot indicated by a red arrow on Fig.~3A)'''. 
     43---- 
     44== Labels == 
    4345 
    44 === Gap filling algorithm === 
     46=== Unlabelled areas === 
    4547 
    46 It is handled by a heuristic gap filling algorithm ([wiki:barSoftwareGapFillDetails detailed description]). The main idea 
    47 behind this algorithm is to expand the contours by applying a [http://en.wikipedia.org/wiki/Mathematical_morphology dilation] filter ([http://www.imageprocessingplace.com/DIP-3E/dip3e_main_page.htm Gonzalez2008]) until the boundary closes. If a very 
    48 accurate reconstruction is required, the best policy is to prepare a 
    49 precise contour slide so there is no need to apply gap filling. For 
    50 the atlases we tested, the gap filling algorithm performed well, the 
    51 reconstructed structures were adequate and the gains in time were 
    52 tremendous as compared with manual cleaning. 
     48Another problem we encountered during tracing was the lack of labels 
     49attributed to some areas.  3dBAR vector parser automatically detects 
     50such unlabeled regions by comparing the sum of traced paths with the 
     51total area of the ''Brain'' structure. The parser locates all 
     52contiguous unlabelled regions consisting of more than a given number 
     53of pixels and traces them.  The resulting structures are called 
     54''Unlabelled'' and may be indexed and reconstructed similarly to 
     55the regular structures. 
    5356 
     57 
     58Left: 
     59 
     60Middle: 
     61 
     62Right: 
     63 
     64 
     65[[Image(001__unlabelled_annotated.gif)]] 
     66 
     67=== Displaced labels === 
     68 
     69Left: 
     70 
     71Right: 
     72 
     73[[Image(003__invalid_labels_annotated.gif)]] 
    5474 
    5575 
    … …  
    80100 
    81101 
    82 === Unlabelled area === 
     102---- 
    83103 
    84 Another problem we encountered during tracing was the lack of labels 
    85 attributed to some areas.  3dBAR vector parser automatically detects 
    86 such unlabeled regions by comparing the sum of traced paths with the 
    87 total area of the ''Brain'' structure. The parser locates all 
    88 contiguous unlabelled regions consisting of more than a given number 
    89 of pixels and traces them.  The resulting structures are called 
    90 ''Unlabelled'' and may be indexed and reconstructed similarly to 
    91 the regular structures. 
     104== Leaking structures and gap filling algorithm == 
    92105 
     106The most common defect is gaps in structure outlines. It frequently 
     107happens when two contour lines, which should touch are drawn so that 
     108they leave a little space in between. Such an arrangement may not be 
     109visible in a printed atlas but it greatly disturbs the tracing process 
     110where every pixel may influence the results. In this case, the 
     111structure being traced overtakes the space of its neighbour through 
     112the broken contour which we call ''leaking of the structure'' '''(see 
     113the spot indicated by a red arrow on Fig.~3A)'''. 
    93114 
     115It is handled by a heuristic gap filling algorithm ([wiki:barSoftwareGapFillDetails detailed description]). The main idea behind this algorithm is to expand the contours by applying a [http://en.wikipedia.org/wiki/Mathematical_morphology dilation] filter  
     116([http://www.imageprocessingplace.com/DIP-3E/dip3e_main_page.htm Gonzalez2008]) until the boundary closes. If a very accurate reconstruction is required, the best policy is to prepare a 
     117precise contour slide so there is no need to apply gap filling. For 
     118the atlases we tested, the gap filling algorithm performed well, the 
     119reconstructed structures were adequate and the gains in time were 
     120tremendous as compared with manual cleaning. 
     121 
     122Left: 
     123 
     124Middle: 
     125 
     126Right: 
     127 
     128[[Image(002__gapfill_annotated.gif)]] 
     129 
     130---- 
     131 
     132== Slice details improvement == 
     133 
     134=== Individual substructures singled out  === 
     135 
     136...by simple line drawing and new regions labeling.  
     137 
     138Unrecognized areas can be attributed to the structures from the hierarchy level.  
     139Here unrecognized thalamic nuclei which were not labeled with their specific labels are all labeled with "Th" so they all will be included in the final reconstruction of the thalamus. 
     140 
     141Left: 
     142 
     143Right: 
     144 
     145[[Image(0041_new_edges_animated.gif)]] 
     146 
     147---- 
     148=== Additional elements apart anatomical data included in CAF slide === 
     149 
     150In our example we include in the slide the outline of the electrode. Space occupied by lesions, or injected dyes 
     151can be equally added to brain elements. 
     152 
     153Left: 
     154 
     155Right: 
     156 
     157[[Image(007__final_electrode.gif)]] 
friendster counter