CN101042771A - Medicine image segmentation method based on intensification learning - Google Patents

Abstract

This invention discloses one medical cut method based on study, which through different image specimens and according to defined shape, operation and credit, processing intense study and environment interacting and adopting error test type best strategy; finally forming one new intense study image cut method by use of study knowledge similar medical images for cutting.

Description

Medical image cutting method based on intensified learning

Technical field

The present invention relates to a kind of based on medical image cutting method intensified learning, increment.

Background technology

To accurately cutting apart of image object be one of important foundation of medical image pattern-recognition.In the application of lung carcinoma cell identification, in the most lung carcinoma cell image between nucleus and the endochylema contrast lower, obscurity boundary between nucleus edge and the background is added the background impurities The noise, and this all makes and is difficult to the lung carcinoma cell image is cut apart more accurately.

Traditional image partition method mainly is divided into based on cutting apart of threshold value and cutting apart based on gradient.The former can accurately not be partitioned into the target area for the image with multimodal grey level histogram.And the latter can not well be partitioned into the target area equally for the approaching situation of target and background gray scale.In addition, because the image acquisition of lung carcinoma cell pathological image has very big otherness, this causes common image partitioning method to be difficult to adapt to complex environment like this.

At present, intensified learning has been widely used in prediction, Based Intelligent Control, numerous areas such as Flame Image Process.Compare with the traditional images dividing method, have the incremental learning ability, can adapt to complex environment, medical science lung carcinoma cell image is made correct cutting apart based on the medical image cutting method of intensified learning.

Summary of the invention

Goal of the invention: the objective of the invention is at the deficiencies in the prior art, provide a kind of and can make the correct medical image cutting method of cutting apart based on intensified learning.

Technical scheme: the present invention is by different image patterns, and according to the state of definition, action and award by learning alternately of intensified learning and environment, adopt the mode of trial and error to learn the optimum behavior strategy.A kind of new image partition method based on intensified learning of final formation utilizes the medical image of the knowledge similar of having learnt to cut apart.This method may further comprise the steps: (1) initialization Q matrix, Q matrix are recorded under current state and all the follow-up states with the two-dimensional array form and go to select to move the accumulation award of doing acquisition with tactful π; (2) the new samples image is carried out sobel (a kind of edge detection operator commonly used) rim detection, obtain edge image; (3) the maximum method of new samples imagery exploitation inter-class variance is cut apart, obtained comprising the bianry image of nucleus and endochylema; (4) definition status is the ratio that current Threshold Segmentation result's objective contour edge and equitant ratio in edge that the sobel rim detection obtains and current Threshold Segmentation result's target area area overlaps with target area area that the maximum method of inter-class variance splits; The definition action increases or reduces action A for current threshold value _iThe gray level of representative, A=[-30-10-5-1 015 10 30]; The definition award is the matching degree that current target area that is partitioned into and image actual optimum are cut apart; (5) calculate the state and the award of each segmentation threshold correspondence of 0-255, so each threshold value and state and award corresponding (6) repeating steps (7), the mean square deviation of 10 times average Q matrix front and back renewal is less than 0.005 up to date; (7) a given initial threshold, the optimum segmentation threshold value is arrived up to changes of threshold in repeating step (8)～(10); (8) obtain current state according to threshold value; (9) adopt ε-greedy strategy (ε-greedy strategy is selected the maximum action of award in the Q-matrix with the probability of 1-ε, selects other actions with the probability of ε) to select action, change threshold value; (10) obtain corresponding feedback award according to the new threshold value after the change threshold value and upgrade the Q matrix; (11) the new samples image is repeated (2)～(10).

Beneficial effect: remarkable advantage of the present invention is: can effectively distinguish nucleus and endochylema, and adaptive, the learning of increment, and accurately image is cut apart.

Description of drawings

Fig. 1 is the frame model of the inventive method.

Fig. 2 is that the module in the inventive method is formed structural drawing.

Fig. 3 is the process flow diagram of the inventive method.

Fig. 4 is that the sobel operator carries out rim detection and obtains edge image.

Fig. 5 is the result images of inter-class variance maximum fractionation.

Fig. 6 is the optimum segmentation result

Embodiment

As shown in Figure 1, the frame model of the inventive method.

As shown in Figure 2, the inventive method comprises state sensing module, action selection module, policy update module, award sensing module and image segmentation module.

The inventive method flow process describes in detail as shown in Figure 3 below:

Step 1, initialization Q matrix, Q matrix are recorded under current state and all the follow-up states with the two-dimensional array form and go to select to move the accumulation award of doing acquisition with tactful π.

Step 2 adopts sobel operator (a kind of edge detection operator commonly used) to carry out rim detection to the new samples image, obtains edge image.The edge detection results image as shown in Figure 4.

Step 3 is carried out the inter-class variance maximum fractionation to the new samples image, obtains comprising the bianry image of nucleus and endochylema.The result images of inter-class variance maximum fractionation as shown in Figure 5.

Step 4, definition status S is the ratio F that current Threshold Segmentation result's objective contour edge and equitant ratio E in edge that the sobel rim detection obtains and current Threshold Segmentation result's target area area overlaps with target area area that the maximum method of inter-class variance splits, i.e. S=(E * F); The definition action increases or reduces action A for current threshold value _iThe gray level of representative, A=[-30-10-5-1 015 10 30]; Definition award R is the matching degree that current target area that is partitioned into and image actual optimum are cut apart.

$E=\frac{\left|{\mathrm{Edge}}_{T}I{\mathrm{Edge}}_S\right|}{\left|{\mathrm{Edge}}_S\right|}---\left(1\right)$

Edge _TBe the current edge of cutting apart, Edge _sEdge for the rim detection extraction.

$F=\frac{\left|{\mathrm{Front}}_{T}I{\mathrm{Front}}_{\mathrm{OSTU}}\right|}{\left|{\mathrm{Front}}_{\mathrm{OSTU}}\right|}---\left(2\right)$

Front _TBe the current target area of cutting apart, Front _OSTUBe the target area of adopting the maximum method of inter-class variance (OSTU) to split.

$R=100\×\frac{\left|B_{O}I{B}_T\right|+\left|F_{O}I{F}_T\right|}{|B_O+F_O|}---\left(3\right)$

B _OBe the background of optimum segmentation, F _OForeground target for optimum segmentation.B _TBe the current background of cutting apart, F _TBe the current foreground target of cutting apart.

Step 5, according to state and award that (1), (2), (3) formula are calculated each segmentation threshold correspondence of 0-255, so each threshold value is corresponding with state and award.

Step 6, repeating step (7), the mean square deviation before and after 10 times the average Q matrix update is less than 0.005 up to date.

Step 7, a given initial threshold, the optimum segmentation threshold value is arrived up to changes of threshold in repeating step (8)～(10).

Step 8 obtains current state according to current threshold value.

Step 9 adopts ε-greedy strategy (ε-greedy strategy is selected the maximum action of award in the Q-matrix with the probability of 1-ε, selects other actions with the probability of ε) to select action, changes segmentation threshold.

Step 10 obtains corresponding feedback award according to the new threshold value after the change threshold value and upgrades the Q matrix.More new formula is a current state suc as formula (4): s, and a is the action of corresponding s, and s ' is the NextState behind the execution action a, and a ' is corresponding s ' action.

$Q(s,a)\←Q(s,a)+\mathrm{\α}[r+\mathrm{\γ}\underset{a^{\′}}{\max }Q(s^{\′},a^{\′})-Q(s,a)]---\left(4\right)$

Step 11 is to new samples image repeating step 2 to 10.

Claims (1)

1, a kind of medical image cutting method based on intensified learning is characterized in that this method may further comprise the steps:

(1) initialization Q matrix, Q matrix are recorded under current state and all the follow-up states with the two-dimensional array form and go to select to move the accumulation award of doing acquisition with tactful π;

(2) adopt the sobel operator to carry out rim detection to the new samples image, obtain edge image;

(3) the new samples image is carried out the inter-class variance maximum fractionation, obtain comprising the bianry image of nucleus and endochylema;

(4) definition status S is the ratio F that current Threshold Segmentation result's objective contour edge and equitant ratio E in edge that the sobel rim detection obtains and current Threshold Segmentation result's target area area overlaps with target area area that the maximum method of inter-class variance splits, i.e. S=(E * F); The definition action increases or reduces the gray level of action Ai representative, A=[-30-10-5-1 015 10 30 for current threshold value]; Definition award R is the matching degree that current target area that is partitioned into and image actual optimum are cut apart;

(5) state and the award of each segmentation threshold correspondence of calculating 0-255, so each threshold value is corresponding with state and award;

(6) repeating step (7), the mean square deviation before and after 10 times the average Q matrix update is less than 0.005 up to date;

(7) a given initial threshold, the optimum segmentation threshold value is arrived up to changes of threshold in repeating step (8)～(10);

(8) obtain current state according to current threshold value;

(9) adopt ε-greedy policy selection action, change segmentation threshold;

(10) obtain corresponding feedback award according to the new threshold value after the change threshold value and upgrade the Q matrix;

(11) to new samples image repeating step 2 to 10.

Images