# US 2004/0001623 A1 · Image analysis process for measuring the signal on biochips

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2004/0001623 A1
- Document ID: US20040001623A1
- Status: Published application
- Kind: A1
- Application number: 10/173,672
- Filed: 2002-06-19
- Published: 2004-01-01
- Earliest claimed: 2002-06-19
- Assignees: Commissariat a l'Energie Atomique
- Inventors: Nicolas Ugolin; Mory Doukoure; Olivier Alibert; Sylvie Chevillard
- Web page: https://patentgrep.com/patent/US20040001623A1

## CPC

- G06T 1/0007: {Image acquisition}
- G06T 7/0012: {Biomedical image inspection}
- G06T 7/11: Region-based segmentation
- G06T 2207/10056: Microscopic image
- G06T 2207/10064: Fluorescence image
- G06T 2207/20056: Discrete and fast Fourier transform, [DFT, FFT]
- G06T 2207/30072: Microarray; Biochip, DNA array; Well plate
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

The invention relates to an image analysis process for measuring the signal on biochips organized into one or several blocks each comprising a large number of spots each composed of at least one probe, the said process being characterized in that it comprises the following steps: a deposit is made with a constant pitch, the spots appearing in the form of a periodic and regular signal; this periodicity property is used to identify the position of the blocks and the spots, this periodicity being searched for on projections of the image in the x and y directions.

## Claims

### Claim 1

Image analysis process for measuring the signal on biochips organized into one or several blocks each comprising a large number of spots each composed of at least one probe, the said process being characterized in that it comprises the following steps:
a deposit is made with a constant pitch, the spots appearing in the form of a periodic and regular signal,
this periodicity property is used to position blocks and spots, this periodicity being required on projections of the image in x and y.

### Claim 2 (depends on 1)

Process according to claim 1, in which averaged projections orthogonal to the x and y axes of the plane of the biochip are used to amplify the periodic signal, and thus two histograms {overscore (H)}j and {overscore (H)}i are obtained.

### Claim 3 (depends on 2)

Process according to claim 2, in which the next step is to use the discrete single dimensional Fourier transform (FFTD) to identify the main frequencies of the image in x and y, the maximum power of the transform P(ω) giving the values ω0, from which the spot repetition periodicities λ0 in x and y are deduced.

### Claim 4 (depends on 3)

Process according to claim 3, in which a variable threshold is applied to the two histograms {overscore (H)}j and {overscore (H)}i, the optimum threshold being given by the value that gives the maximum value of the equation P(ωmax)/
30 P ⁡ ( ω max ) / ∑ i = 10 50 ⁢ P ⁡ ( ω ⁢ ⁢ i ) ,
the distribution of the periodicity λ then being examined along the two histograms {overscore (H)}j and {overscore (H)}i, a window with a size 2 λ0 sliding along the two histograms, and in which the variations of the power of the Fourier transform of these histograms are explored locally, regions in which there is a loss of power corresponding to regions between blocks, and this method supplies a global breakdown of the image.

### Claim 5 (depends on 4)

Process according to claim 4, in which a window with a width of λ0 and a length of 2λ0 centered on the global cut-out row is used to produce a fine delimitation of the boundaries of each block, and the presence of excessively intense and therefore suspicious signals can then be explored locally, and the limit of the blocks around the signal may be redefined if necessary, and in which this is achieved by firstly extending the window by λ0/2 on one side of the limit and then on the other side, and in which if the signal increases only for one of the two extensions of the window, the limit of the block is shifted by λ0 into the adjacent block for which there was no increase in the signal, and in which if there was no increase in the signal or if the signal was increased for the two shifts, then either the signal was a parasite signal or the two blocks are very close,
in which in both cases the probability of the signal belonging to the right and to the left of the cut line is calculated,
and in which P(ω) is obtained for each half projection with and without a suspicious signal, the gain then being equal to:
31 Gain = P ⁡ ( ω + signal ) P ⁡ ( ω - signal )
in which, if the gain is greater than 1 for one of the two half projections only, the signal considered is assigned to it, by shifting the cut line by λ0/2 into the adjacent block; if the gain is not greater than 1 for either of the two blocks, then it is a parasite signal and the cut line is not shifted; if the gain is greater than 1 for both half projections, the signal is in phase with the two half projections and therefore the blocks are very close and in phase, and the signal is assigned to the block for which the gain is highest, by shifting the cut line by λ0/2 into the adjacent block.

### Claim 6 (depends on 1)

Process according to claim 1, in which unilluminated spots are isolated as follows:
the histograms {overscore (H)}j and {overscore (H)}i are made for the rows and the columns of a block respectively,
a FFTD transform is applied to the two histograms in each block,
in which the maximum value of P(ω) gives the periodicity λ0 of the spot repetition,
and in which the least squares method is used to align projections with a sine curve with the following equation:
A sin(ω0Xi+φ)
32 where : ⁢ - ω 0 = 2 ⁢ π λ 0
A: average amplitude of the projections,
φ: phase shift to be determined.

### Claim 7 (depends on 6)

Process according to claim 6, in which a digital reconstruction algorithm is then applied to these histograms, which is used to start from the initial projections and recreate a signal with almost identical amplitudes, the use of a threshold then making it possible to produce a binarized representation of the created history and to localize the precise positions of the spots present.

### Claim 8 (depends on 7)

Process according to claim 7, in which the binarized projection is convoluted using a sinusoidal function of λ0 such as:
33 H ~ i = ∑ j = 0 λ ⁢ ⁢ 0 / 2 ⁢ ⁢ ( A ⁢ ⁢ sin ⁢ ⁢ ( ω ⁢ ⁢ time ⁢ ⁢ j ) ⁢ xHb i + j - λ / 4 )
and in which searching for the minimum and maximum values of the convolution function makes it possible to automatically rebuild a grid surrounding all detectable spots.

### Claim 9 (depends on 1)

Process according to claim 1, in which several images with low excitation and low detection are created and accumulated into a single image.

### Claim 10 (depends on 1)

Process according to claim 1, in which a high resolution image is created and the resolution of the image is then reduced artificially by replacing n adjacent pixels by the average of the n pixels which thus reduces the size of the image by a factor of n, and a homogeneity filter is applied to eliminate the pixel with the greatest heterogeneity with regard to the adjacent pixels.

### Claim 11 (depends on 1)

Process according to claim 1, in which pixels within each spot are sorted to identify the different types of pixels present: background noise, hybridizing signal, parasite signal.

### Claim 12 (depends on 1)

Process according to claim 1, in which a modified EM algorithm is applied in which the classes are made smaller and the concept of a pixel that does not belong to any class is introduced, by applying a threshold to the calculated probability that it belongs to one of the two or three main classes.

### Claim 13 (depends on 12)

Process according to claim 12, comprising the following steps:
an initializing step, which consists of building up a blurred distribution table Ci(k) of the pixels,
an estimating step in which the new blurred distribution table is calculated,
a maximizing step, in which the mix parameters Om are recalculated,
and in which these last two steps are reiterated until convergence of the log-probability classification.

### Claim 14 (depends on 13)

Process according to claim 13, in which a new convergence estimator is introduced based on the similarity or the calculated distance between the real distribution of pixels and the recalculated distribution using the mix parameters.

### Claim 15 (depends on 1)

Process according to claim 1, comprising the following steps:
an image filter step to reduce background noise, by very slightly reducing the significant signal and by increasing the signal to noise ratio, for the image,
a degrading filter step increasing the image contrast between the background noise and the hybridizing signal,
a step for positioning the blocks on the image of the biochip starting from external data about the number of blocks in rows and in columns,
a step to automatically determine the position of each spot, including the position of spots for which there was no signal because there was no hybridizing within its block,
a step in which the spot is automatically broken down into segments to determine pixels representative of the reaction and to calculate an evaluator of the spot signal by the average, the median, etc.

### Claim 16 (depends on 15)

Process according to claim 15, in which a method based on identification of the local loss of the periodic signal is used during the initialization step, such as the power of the Fourier transform of the spot projection signal onto the axes by rows or by columns.

### Claim 17 (depends on 15)

Process according to claim 15, in which a first filter is used during the filter step to rectify the image, if necessary, by alignment of the spots in the same block.

### Claim 18 (depends on 15)

Process according to claim 15, in which a second filter is used during the filter step to amplify the signals specifically originating from the spots and to level other types of signals.

### Claim 19 (depends on 15)

Process according to claim 15, in which the position of each block in the image is refined.

### Claim 20 (depends on 15)

Process according to claim 15, in which the hybridizing pixels are broken down into segments using a EM method.

### Claim 21 (depends on 20)

Process according to claim 20 in which a constraint in the EM method used is relaxed, in other words pixels too far away from the assumed classes are not assigned.

### Claim 22 (depends on 15)

Process according to claim 15, in which a morphomathematical method defines the objects present in the frame of the spot to initiate segmentation classes.
FIG. 23a
Initial image
Non-degrading filter
Non-degraded filtered image
Degrading filter
Degraded filtered image
Global positioning of blocks
Refinement of block positioning
Intensity evaluator for each spot (average, median, etc.)
Segmentation of the pixels for each spot
Spot position grids
Delimitation of spots
FIG. 23b
Several initial images of the same slide for different markers
Non degrading filter
Non degraded filtered images
Degrading filter
Degraded filtered images
Superposition of images
Superposed degraded filtered image
Global positioning of the blocks
Intensity evaluator for each spot (average, median, etc.)
Segmentation of pixels for each spot (EM with or without morphomathematics)
For each image: position of the grid taking account of the offset of the superposition
Spot position grids on the superposed image
Delimitation of spots
Refinement of the position of blocks
FIG. 23c
Initial image
Global position of blocks
Refinement of the position of blocks
Calculation of an intensity evaluator for each spot
Segmentation of pixels for each spot
Spot delimitation grids
Spot delimitation
