Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/60—Analysis of geometric attributes
  • G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/30—Subject of image; Context of image processing
  • G06T2207/30004—Biomedical image processing
  • G06T2207/30061—Lung

Definitions

• the present invention relates to a method for the automated analysis of tissue morphology.
• the method of the invention relates to a computer-based analysis of lung tissue, for the detection of, for example, pulmonary emphysema.
• the surface area of the lung is approximately 130 m 2 , which is packed into the limited space of the chest cavity (around 5-6 litres).
• the internal lung surface is a highly folded membrane that allows the organ to be very efficient at gaseous exchange, with around 300,000,000 alveoli all open to outside air (1).
• This relationship of lung structure to function is highly conserved and a change in respiratory function will often be reflected in lung structure. For this reason lung morphometry has been widely used in the study of lung disease and dysfunction. Examples include bronchopulmonary dysplasia (2), pulmonary fibrosis (3) and pulmonary emphysema (4).
• Pulmonary emphysema is defined as abnormal and permanent enlargement of the airspaces distal to the terminal bronchioles, accompanied by destruction of their walls without obvious fibrosis (5).
• Examination of thin slices from emphysematous lungs reflects this definition ( Figure 1), and the severity of emphysema obseived from such slices, has been commonly evaluated by calculating the average distance between alveolar walls (Lm) (6,7).
• Lm alveolar walls
• An increasing Lm value reflects increasing airspace size which in turn, is a feature of more severe emphysema (8).
• Thin slices of tissue are two-dimensional but the micro structures that they sample are tliree-dimensional.
• Stereological techniques allow inferences about tliree-dimensional geometric properties of these structures to be made from measurements made on two- dimensional slices.
• Various stereological techniques for lung structure morphometry have been described, including volume estimation by the Cavalieri method (9), and surface area to volume ratio estimation using a graticule (10). Structure specific measurements such as number and volume of alveoli have also been described (11).
• Emphysema in man has also been examined using stereological techniques on thin sections of lung after autopsy (12).
• Various animal models of emphysema exist which provide the opportunity for potential therapeutic agents to be evaluated (13), and morphological techniques have been employed to address the validity of these models.
• a method of analysing a sample image representing a sample of lung comprising: forming an applied graticule image corresponding in size to said sample image and excluding areas corresponding to non-parenchyma within said sample image; superimposing said applied graticule image on said sample image; deteraiining one or more parameters dependent upon which portions of said applied graticule image fall in airspace and which portions of said applied graticule image fall upon tissue; and calculating a parameter indicative of a surface to volume ratio for said sample of lung from said one or more parameters.
• sample image is an image of lung tissue which is obtained by the image analyser and used to perform the morphometric analysis as described herein.
• the sample image may be of the whole lung or part of the lung. Typically, it is of part of then lung.
• the sample image is advantageously a composite image, made up of two or more images of lung tissue taken from different parts of the lung.
• a "lung sample” is the part of the lung which is used to obtain the sample image. It may be a section of lung, a tissue specimen or biopsy, or a whole lung, and may be imaged in vitro or in vivo.
• a "graticule” is a grid, mesh or other means to divide an image area into a plurality of sections, which may be regular or irregular in shape.
• the graticule divides the image into a series of regular sub-sections, for example squares.
• Airspace means the space at the interior of alveoli or other air-filled spaces in the lung. In a tissue section, it refers to the gaps between tissue where air would be present in vivo, regardless of whether air is so present in the section itself as analysed.
• the applied graticule image is formed by the steps of: identifying non-parenchyma areas of said sample image corresponding to tissue other than parenchyma; forming a full frame graticule image corresponding in size to said sample image; and removing from said full frame graticule image areas corresponding to said non- parenchyma areas to form said applied graticule image.
• the step of identifying comprises the steps of: searching said sample image for areas of tissue greater than a predetermined size; classifying said areas of tissue greater than a predetermined size as non- parenchyma areas.
• User input may also be exploited to improve the identification procedure.
• the parameters measured in the method of the invention preferably include one or more of: a number I a of graticule lines of said applied graticule image intersecting tissue; and a number P a of ends of graticule lines on airspace.
• Lx represents a length of a graticule line
• the invention provides a method of analysing a sample image representing a sample of lung, said method comprising: identifying non-parenchyma areas of said sample image corresponding to tissue other than parenchyma; searching within areas of said sample image corresponding to parenchyma for bounded areas of airspace surrounded by tissue; measuring a perimeter value and an area value for said bounded areas; and calculating a parameter indicative of a surface to volume ratio for said sample of lung from said perimeter value and said area value.
• Parenchyma can be identified as set forth in the foregoing aspect of the invention. Preferably, bounded areas having an area smaller than a predetermined threshold value are excluded from the calculating step.
• perimeter value can be a total perimeter value P ⁇ for said bounded areas upon which said calculation is based and said area value can be a total area value A T for said bounded areas upon which said calculation is based.
• the surface to volume ratio can thus be calculated as being proportional to:
• the invention provides a method of analysing a sample image representing a sample of lung, said method comprising: forming an applied lines image of a plurality of substantially parallel lines and corresponding in size to said sample image and excluding areas corresponding to non- parenchyma within said sample image; superimposing said applied lines image on said sample image; determining where lines of said applied lines image intersect tissue within said sample image to form intersected line segments; and calculating a parameter indicative a mean length of said intersected line segments.
• the steps of forming, superimposing, determining and calculating can be repeated for one or more further applied lines images of lines running in a different direction.
• the different direction can be substantially orthogonal.
• the method further comprises calculating a parameter indicative of a distribution of lengths of said intersected line segments.
• said applied lines image is formed by the steps of: identifying non-parenchyma areas of said sample image corresponding to tissue other than parenchyma; fo ⁇ ning a full frame lines image corresponding in size to said sample image; and • removing from said full frame lines image areas corresponding to said non- parenchyma areas to form said applied lines image.
• Parencyma and non-parencyrna can be identified as set forth above.
• a method of analysing a sample image representing a sample of lung comprising: identifying non-parenchyma areas of said sample image corresponding to tissue other than parenchyma; searching within areas of said sample image corresponding to parenchyma for nodes corresponding to ends of branches in tissue of said sample lung; calculating a parameter indicative of a straight line distance between nodes at ends of branches of tissue of said sample of lung; and calculating a parameter indicative of a distance measured along said branches between nodes at ends of branches of tissue of said sample of lung.
• Parenchyma and non-parenchyma can be identified as above.
• areas of tissue within said sample image are thinned prior to said step of searching.
• the steps of searching include applying a convolution to said sample image.
• the nodes are removed from said sample image to leave disconnected branches o f tissue .
• the invention provides a method of analysing a sample image representing a sample of lung, said method comprising: analysing said sample of lung according to the methods set forth in any of the four preceding aspects of the invention to allow two or more of Linear Mean Intercept (LMI), Surface Area to Volume (S/V) ratios, and mean branch length to be calculated; and combining the results of said methods to produce a measure of lung emphysema.
• LMI Linear Mean Intercept
• S/V Surface Area to Volume
• the method according to the fifth aspect of the invention combines all four of the preceding aspects of the invention.
• the invention further provides a computer program product including a computer program for controlling a computer to perform a method as described in the foregoing aspects of the invention, as well as an apparatus for analysing a sample image representing a sample of lung, said apparatus comprising data processing logic operable to perform data processing operations in accordance with a method as described in the foregoing aspects of the invention.
• FIG. 1 Thin sections from human lung illustrating the morphological difference observed when comparing normal to emphysematous lungs. Low power light microscopy images of human lung obtained from thin sections. Normal lung (A) contains regular size alveoli and alveolar ducts, whereas emphysematous lungs present damaged alveoli and airspace expansion (B).
• Figure 2 Image showing the mosaic macro in progress. Images are captured in a raster fashion and each field of view is saved separately. In addition a reference mosaic image (above) is generated which shows the whole area of capture. Typically between 150 and 300 field of view images are required for a whole slice from an adult rat to be captured.
• Figure 3 Single field of view image captured during the mosaic process.
• Figure 4. 3x3 image generated from 9 images captured during the mosaic process.
• FIG. Automated detection of blood vessels and other solid tissue components (shown in green, with original 3x3 as reference).
• Figure 9 Remaining non-parenchyma components identified manually by outlining in green on the image using a digital pen by the user.
• Figure 10 Non-parenchyma component 'b' image.
• Figure 11 A multipurpose test grid first proposed by Weibel for calculation of S/V Ratios (25).
• Figure 13 Image showing part of the grid when merged with the 'b' image - removing lines that would cover non-parenchyma components.
• Figure 18 One pixel thick lines 10 pixels apart drawn on a blank 3x3 and then merged with the 'b' image to leave lines that will only lie over parenchyma.
• Figure 20 Overlay of the full LMI grid on a lung section.
• FIG 21 Thinning of the lung structure resulting in a skeleton- like appearance of the lung parenchyma.
• Original lung binary image (A) is thinned and then smoothed leaving a skeleton stmcture (B).
• Figure 22 Branch structure overlaid on top of original lung binary image (individual branches shown in different colours).
• FIG 23 Effect of PPE on lung morphology. Airspaces are greatly enlarged after PPE treatment (B) resulting in emphysematous morphology compared to that of normal rat lung (A).
• FIG 24 Surface to Volume ratios of adult rat lungs following a single instillation of saline or PPE and examined 28 days later. S/V ratios were calculated by using a Weibel graticule (A) or perimeter/area based measurements (B) (see methods).
• Figure 26 Normalised distribution of intercept lengths from LMI analysis. PPE treatment causes a decrease in small and an increase in large intercepts.
• Figure 27 Significant effect of PPE on mean node-to-node distance and mean branch length dete ⁇ nined by nodal analysis (see methods).
• Figure 28 Normalised distribution of branch lengths from node analysis. PPE treatment causes a decrease in small branches and an increase in large branches.
• FIG. 29 Relationship between S V ratio (Weibel Graticule with 15 pixel lines) and LMI in all rats examined. S/V ratio is inversely proportional to LMI.
• Figure 30 Flow chart of the computer-implemented method of the invention.
• Porcine pancreatic elastase (PPE) 600 units in 0.5 ml), or an equal volume of saline (to serve as controls), was instilled transorally into the trachea of anaesthetised adult (250- 300g) male Spraque-Dawley rats using a Penn Century micro-sprayer. The rats were monitored for 24 hrs for signs of distress and approximately 25% had to be sacrificed due to the severity of their reaction to the elastase. The remaining animals were killed 28 days later for assessment of emphysema.
• PPE Porcine pancreatic elastase
• the rats were killed with an overdose of sodium pentobarbital (-0.5 ml of a 0.8 M solution i.p.).
• the diaphragm was punctured after intubation of the trachea and the lungs inflated with phosphate-buffered formalin (PBF) at a transpulmonary pressure of 20cm H 2 0.
• PPF phosphate-buffered formalin
• the trachea was li gated, the lungs removed from the thorax and stored in PBF for 48 hours at 4° C.
• the left and right lungs were separated and a longitudinal section ⁇ 7 mm thick was cut from the middle of the left lobe. This was dehydrated through an ethanol gradient and embedded in paraffin wax using a TissueTek processor. After facing the block, 4 ⁇ m sections were cut and mounted on slides.
• the wax was removed with xylene and the tissue stained with haematoxylin and eosin (H&E).
• a Leica DMR microscope fitted with a xlO objective was used to view the lung slides.
• a JVC 3-CCD colour video camera (RGB-PAL) was fitted to the microscope to allow image capture.
• a Prior motorised stage allowed computer controlled navigation of the field of view in the x and y axis.
• Image capturing and analysis was carried out by Carl- Zeiss KS400 image analysis software (Imaging Associates, Thame, UK) on a Compaq Deskpro EN series PC (256 MB of RAM, Intel Pentium III 600 MHz processor) rui ing Microsoft Windows NT 4.
• the KS400 software contains a command line interpreter, enabling user-defined macros to be written and implemented. Macros referred to in the text are presented in the appendix.
• a longitudinal whole lobe slice from an adult rat typically covers a large area of a standard microscope slide; especially if the lung has been exposed to PPE and the lung is enlarged. Images of the lobe are captured using the "Mosdefme" macro ( Figure 2). The four corners of the area to be scanned (the whole lobe slice) are defined by moving the field of view to these positions and the co-ordinates are automatically recorded. Under computer control the camera grabs an image of the current field of view and then the stage moves the slide so that the next field of view is adjacent to the previous one. This process is continued in a raster fashion until the whole lobe is captured. Typically between 150 and 300 images are grabbed for an adult rat lobe slice. Grey rather than colour images are saved to preserve hard disc space. The lung tissue appears dark compared to the airspace ( Figure 3). The final on-screen magnification using the xlO objective is approximately x300.
• 3x3 generation, segmentation, non-parenchyma component identification and image storage are achieved in one process by running the 'RatXlOproc' macro. However this process has been broken down into separate components below for clarity.
• Lung tissue is made up of many components such as alveoli, alveolar ducts, blood vessels, airways, bronchioles and airway ducts.
• Non-parenchyma components such as blood vessels and conductive airways are not randomly distributed in the lung (24) and therefore have the potential to influence the analysis. (Moreover, a large airway will be indistinguishable from a large emphysematous area from an analytical viewpoint). For these reasons non-parenchyma components are omitted from the analysis. This process is achieved in two stages: an automated step followed by a manual step.
• Blood vessels and other large areas of solid tissue can be identified automatically due to the size of the area they occupy. These areas are then projected on top of the 3x3 in green using an overlay (Figure 8).
• This automatic step identifies tissue-based components but it cannot be used to identify large airways, as these are indistinguishable from large airspaces. Therefore other non-parenchyma components not picked up by the automated step are identified manually using a digital pen by the user ( Figure 9).
• Features to be omitted from the analysis are stored as a new image (termed 'b' image) and this is created from the identified components previously illustrated in green ( Figure 10).
• This procedure is completed for each of the six 3x3 images.
• the analysis process (detailed below) is fully automated, it is preferable to perform the 3x3 image generation process for other slides and build up a 'bank' of images to be analysed. The analysis can then be run overnight, for example.
• the Surface Area to Volume ratio can be calculated from point counting measurements made using a Weibel graticule ( Figure 11).
• the graticule is normally used either through the microscope eyepiece, or projected onto micrographs of lung sections.
• the grid test lines probes
• the grid test lines give an indication of the surface area by recording the number of intersections between airspace and tissue. The ends of these lines serve as references for volume.
• Using the inverse of a formula for Volume to Surface Area ratio (25) we have adapted this methodology to our computerised system. We decided to superimpose the graticule on the whole of our 3x3 image to give a large number of test lines/image.
• the Weibel graticule is made up of short test lines (length d) whose end points are arranged in a regular equilateral triangular lattice resulting in a rhombus lattice unit ( Figure 12).
• d short test lines
• Figure 12 a regular equilateral triangular lattice resulting in a rhombus lattice unit
• both the length of the line d and the value resulting from d/2 • 3 need to be integers, We empirically determined 15 and 30 as values for d, which gave integers for m as 13 (12.99), and 26 (25.98) respectively.
• the macro draws a line in the top left hand corner of a blank 3x3 image d pixels long in the x axis, moves along a further d pixels and draws another one. This is repeated until the right edge of the image is reached when the y position is incremented m pixels down and the starting x position becomes 0 + d/2. The lines are then drawn as before until the right edge of the image is reached.
• the y position is incremented again m pixels down and the x position is reset to zero.
• the resultant image now contains a Weibel graticule, which covers the entire area of a 3x3.
• Both 15 and 30 pixel length lines are used, as both have (at x300 magnification) a suitable length relative to the features of interest such as alveoli.
• the formula for calculating the S/V ratio is: l_ x 2 / P a x L T (25), where I a is number of lines intersecting the tissue (I ⁇ - _), and LT is the length of an individual line on the grid (d).
• I a is number of lines intersecting the tissue (I ⁇ - _)
• LT is the length of an individual line on the grid (d).
• the S/V ratio can be directly determined from the image by perimeter and area measurements.
• Weibel (25) the S/V ratio is a function of the perimeter divided by the area. Perimeter and area measurements are carried out automatically, with each individual area being indicated with a different colour on the graphic display ( Figure 16). For each area the perimeter is also measured, and this results in a data list containing area and perimeter measurements for each region. The 'b' image is used to prevent the non-parenchyma components being measured as before. To prevent noise interfering with the data, any region with an area less than 100 pixels is not measured. To arrive at a S/V ratio, the following formula is used: 4 x Total Perimeter / ⁇ x Total Area.
• LMI Linear Mean Intercept
• the lines for the LMI were overlaid on the whole 3x3, first in the x direction, parallel lines 10 pixels apart, and then merged with the 'b' image to leave lines only covering parenchyma (Figure 18).
• This parenchyma grid is then merged with the lung 3x3 ('a' image) resulting in an image containing many intercepts ( Figure 19).
• Each individual intercept longer than 10 pixels is measured and the corresponding value entered into the data list automatically.
• the grid is then drawn in the y-axis, merged with the ' b' and 'a' images and the measurements repeated.
• This process results in the lung image being analysed by a LMI grid containing many lines ( Figure 20).
• the data list typically contains several thousand values. This procedure is repeated for the other 3x3 's and the LMI is calculated as the mean of all values measured.
• the data also allows the distribution of the intercept length to be examined.
• the S/V ratio (determined by either of the above methods) is influenced by changes in tissue architecture and airspace enlargement.
• the LMI measurements come solely from airspace measurements.
• the forth analysis parameter, nodal analysis examines the tissue component of the lung.
• the binary lung structure image ('a' image) is first merged with the 'b' image to leave only lung parenchyma.
• the lung structure is then thinned down to one-pixel thick lines while preserving the lung skeleton, and smoothed to remove artefacts (Figure 21).
• Using a convolution filter the branching points are detected and then removed leaving only individual branches of the lung structure on the image ( Figure 22).
• the straight-line distance between the two ends of a branch (node-to-node distance) and the actual branch length (taking into account any curvature of the branch) are measured.
• the S/V ratio determined by perimeter/area measurements was also significantly different between the two groups.
• the S/V ratio was 798 cm “1 ⁇ 96 compared to 1517 cm "1 ⁇ 141 with saline.
• lung morphometric techniques described by others (6,7,25) to carry out computer-based image analysis of thin sections of lung presenting normal and emphysematous morphology.
• Lung morphometry is an inherently labour intensive process but using computer automation techniques we have greatly reduced the time required to obtain quantitative data from such studies.
• the morphometric parameters that we have employed significantly detect emphysematous morphology in an elastase-induced model of emphysema.
• Image capturing using the mosaic methodology described here, has several advantages. Firstly, it is quick, taking between 5-10 minutes on average for each lobe, which can result in 150-300 fields being digitally captured and stored. This procedure is fully automated and requires little training for the user. Secondly, the mosaic process allows large areas of lung tissue to be analysed (and thus a larger number of observations made) by joining up adjacent fields to form larger images such as a 3x3. The Prior motorised stage allows precise movements ( ⁇ 1 ⁇ m accuracy) to be made, enabling these 3x3 images to be created automatically.
• Lm and LMI from the same images will not be identical, as the Lm does not take into account alveolar wall thickness.
• a line 100 ⁇ m in length intersected 4 times would give a Lm of 25 ⁇ m.
• the LMI however would be lower than this as part of the line would fall on tissue and so the actual length falling on airspace would measure perhaps 80 ⁇ m, resulting in an average LMI of 20 ⁇ m.
• Escolar et al. have used a computer-based technique similar to our LMI called alveolar chord length (23) which they measured alongside the conventional Lm. The alveolar chord length was indeed lower than the Lm from the same images.
• Age also has a prominent effect on lung dimensions.
• alveolar dimensions and thus measurements such as the Lm
• this effect has been given the term 'the senile lung' (33), which presents significant airspace enlargement but is distinct from emphysema due to the lack of alveolar destruction.
• the Lm significantly differed by 12 ⁇ m between adulthood (16 weeks old) and middle age (56 weeks old) (34).
• the final factor affecting Lm values across studies is that of shrinkage due to fixation, dehydration, embedding, and sectioning. Correction factors for these components are calculated by taking measurements of volumes or lengths during processing of the lung tissue so that the dimensions of the fresh lung can be estimated (27). Shrinkage was not corrected for in the study reported here. However, it is assumed that as all samples were processed together, shrinkage would be consistent.
• the Lm and LMI can be compared (although the difference between groups with LMI will be slightly larger as alveolar thinning does take place in emphysematous lungs) to approximate the severity of emphysema with different treatments.
• the PPE protocol in this study was based on that used by Massaro et al. in a study demonstrating reversal of elastase-induced emphysema by retinoic acid (35).
• the Lm increased from 74 to 96 ⁇ m compared to our LMI, which rose from 53 to 93 ⁇ m suggesting that the emphysema was more extensive in our study.
• the lungs were inflated to the same pressure, fixed and processed in a similar manner to our study, but the magnification used was xlOO rather than x300 (36). The researchers also made corrections for slrrinkage in this study. The S/V ratio for these rats was 305 cm "1 . Another group carried out almost the same study but with rats of a slightly older age (37) and estimated an S/V ratio of 641 cm “1 . These results illustrate the variation in absolute values obtained using this method.
• lung volume (9,38,26)
• the techniques described here provide a rapid, semi-automated image analysis procedure for obtaining accurate quantitative data to detect and quantify the degree of emphysema in animal models. This procedure, in tandem with lung function measurements, is suitable for use in the evaluation of therapeutic agents intended to prevent or reverse pathological features of emphysema.

Landscapes

• Physics & Mathematics (AREA)
• Engineering & Computer Science (AREA)
• Geometry (AREA)
• Computer Vision & Pattern Recognition (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Apparatus For Radiation Diagnosis (AREA)
• Measuring And Recording Apparatus For Diagnosis (AREA)
• Investigating Or Analysing Biological Materials (AREA)
• Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Priority Applications (3)

Applications Claiming Priority (2)

Related Child Applications (1)

Publications (1)

Family

ID=9955984

Family Applications (1)

Country Status (5)

Families Citing this family (7)

Citations (2)

Family Cites Families (2)

• 2003
  • 2003-04-01 GB GBGB0307537.1A patent/GB0307537D0/en not_active Ceased
• 2004
  • 2004-03-29 JP JP2006506042A patent/JP2006521854A/ja active Pending
  • 2004-03-29 EP EP04724045A patent/EP1611549A1/en not_active Withdrawn
  • 2004-03-29 WO PCT/GB2004/001348 patent/WO2004088593A1/en not_active Ceased
• 2005
  • 2005-09-29 US US11/239,499 patent/US20060093195A1/en not_active Abandoned

Patent Citations (2)

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events