阅读说明：本技术 一种医学ct图像存储与检索方法 (Medical CT image storage and retrieval method ) 是由 刘杨 刘子超 于 2020-06-17 设计创作，主要内容包括：本发明涉及一种医学CT图像存储与检索方法,属于图像存储与检索技术领域。本发明获取若干医学CT图像组建数据库,对其进行压缩、计算DCT矩阵等环节,生成64位哈希指纹；接着对哈希指纹建立多位索引进行存储；对待检索医学CT图像也生成64位哈希指纹,根据多位检索算法在数据库中检索与待检索医学CT图像的哈希指纹相似的哈希指纹,将其所对应的医学CT图像、标签属性等信息输出。本发明与现有技术相比,主要解决了现有技术针对海量医学CT图像相似检索时存在的效率不高、有疾病界限等现象,在保持相似精度的同时优化了检索效率,实现了跨疾病相似检索。(The invention relates to a medical CT image storage and retrieval method, and belongs to the technical field of image storage and retrieval. The method comprises the steps of acquiring a plurality of medical CT image building databases, compressing the medical CT image building databases, calculating DCT matrixes and the like, and generating 64-bit hash fingerprints; establishing a multi-bit index for the hash fingerprint for storage; and generating 64-bit hash fingerprints for the medical CT image to be retrieved, retrieving the hash fingerprints similar to the hash fingerprints of the medical CT image to be retrieved in the database according to a multi-bit retrieval algorithm, and outputting the corresponding medical CT image, label attribute and other information. Compared with the prior art, the method mainly solves the problems of low efficiency, disease boundary and the like existing in the prior art when the similarity search of massive medical CT images is carried out, optimizes the search efficiency while keeping the similarity precision, and realizes the cross-disease similarity search.)

1. A medical CT image storage and retrieval method is characterized in that:

step 1: obtaining a medical CT image img_i,i∈[1,D]Constructing a database, wherein D is the number of medical CT images img_i,i∈[1,D]Label with label attribute_i,i∈[1,D]I.e. medical CT image img_i,i∈[1,D]The corresponding disease name;

step 2: for medical CT image img in database_i,i∈[1,D]Compressing and calculating DCT matrix to generate 64-bit Hash fingerprint hash_i,i∈[1,D]The Hash fingerprint can uniquely represent the img of the medical CT image_i,i∈[1,D]；

Step 3: hash 64-bit fingerprint hash_i,i∈[1,D]Establishing a multi-bit index for storage;

step 4: generating a 64-bit Hash fingerprint rethrash for the medical CT image to be retrieved by the restig according to the Hash fingerprint generation process in Step 2;

step 5: searching the hash fingerprint hash similar to the hash fingerprint rethrash of the medical CT image rettimg to be searched in the database according to a multi-bit search algorithm_i,i∈[1,D]If the Hamming distance N of the two Hash fingerprints is less than or equal to alpha, the Hash fingerprints are hashed_i,i∈[1,D]Corresponding medical CT image img_i,i∈[1,D]Label attribute label_i,i∈[1,D]And (6) outputting.

2. The medical CT image storage and retrieval method of claim 1, wherein: in Step1, the medical CT images include normal medical CT images and patient medical CT images, and the number of the medical CT images is uniformly distributed.

3. The medical CT image storage and retrieval method of claim 1, wherein: in said Step2, the medical CT image img_i,i∈[1,D]The generated hash fingerprint hash_i,i∈[1,D]The specific implementation steps are as follows:

step2.1: img medical CT image_i,i∈[1,D]Is compressed to 32 x 32, or cut intoA small square of size 32 x 32;

step2.2: img the compressed medical CT image_i,i∈[1,D]Converted into 256-step gray scale image and expressed as matrix I_i,i∈[1,D]In which I_i(m,k),i∈[1,D],m∈[1,32],k∈[1,32]Is a matrix element value;

step2.3: calculating the matrix I_i,i∈[1,D]DCT matrix T_i,i∈[1,D]Where the matrix element values are denoted T_i(m,k),i∈[1,D],m∈[1,32],k∈[1,32]；

Step2.4: preserving DCT matrix T_i,i∈[1,D]The upper left corner of the low-frequency part is 8 multiplied by 8, and the rest part is deleted;

step2.5: computing a DCT matrix T_i,i∈[1,D]Average value u of_i,i∈[1,D]Will DCT matrix T_i,i∈[1,D]According to the rule that m is 1 → 8, and k is 1 → 8 from left to right from top to bottom, the medical CT image img is generated by performing binarization according to the formula (1) and connecting the binarizations_i,i∈[1,D]64-bit hash fingerprint hash_i,i∈[1,D]；

4. The medical CT image storage and retrieval method of claim 1, wherein: in Step3, the multi-bit index refers to hash of the fingerprint_i,i∈[1,D]Segmentation is performed according to M segments, as shown in formula (2), where M > α, hash_i,j,i∈[1,D],j∈[1,M]Representing a hash_i,i∈[1,D]Then hash the fingerprint, and_i,i∈[1,D]m segment sub-fingerprint hash_i,j,i∈[1,D],j∈[1,M]Performing (M-alpha) bit permutation and combination to establish an inverted index table, namely, hash of each hash fingerprint_i,i∈[1,D]Will have

hash_i＝[hash_i,1,hash_i,2,…,hash_i,M],i∈[1,D](2)

Technical Field

The invention relates to a medical CT image storage and retrieval method, and belongs to the technical field of image storage and retrieval.

Background

In disease diagnosis, as imaging devices become complicated, the number of images for detecting diseases is increasing, resulting in an increasing workload on doctors. With the development of big data and artificial intelligence, in a cloud computing environment, the rapid retrieval of similar medical CT image characteristics in a large number of medical CT images has great significance for the diagnosis of diseases.

In the case of medical CT images, different cases often lead to different features of the medical CT images, but similar medical CT image features have a strong similarity among similar cases. The prior art discloses a medical CT image storage and retrieval method of random forest hash, which is applied for 201810757397.7, the technology obtains a random forest hash model by training on a medical CT image set, and stores the model and a hash code library corresponding to an image library; when a user inputs a new image to be retrieved, firstly, the model maps the image into a hash code; then searching K nearest hash codes in a hash code library; and finally, decoding and reconstructing the K hash codes into an image by utilizing the maximum consistent rule defined by the decision path of the tree and returning the image to the user. Although the technology can search for medical CT images, the technology uses a decision tree to generate hash codes, which inevitably causes unqualified searching efficiency during searching, and in order to improve the efficiency of image diagnosis, a technology capable of quickly searching the names and treatment information of similar cases in the past is still needed so as to provide reference for doctors to diagnose the disease conditions.

Disclosure of Invention

The invention aims to solve the technical problems of limitation and deficiency of the prior art, and provides a medical CT image storage and retrieval method to solve the problems of low efficiency, disease limitation and the like existing in the prior art aiming at similar retrieval of massive medical CT images.

The technical scheme of the invention is as follows: a medical CT image storage and retrieval method comprises the following specific steps:

step 1: obtaining a number of medical CT images img_i,i∈[1,D]Constructing a database, wherein D is the number of medical CT images, and the img of the medical CT images_i,i∈[1,D]Should have the label attribute label_i,i∈[1,D]I.e. medical CT image img_i,i∈[1,D]The corresponding disease name.

Step 2: for medical CT image in databaseimg_i,i∈[1,D]Compressing and calculating DCT matrix to generate 64-bit Hash fingerprint hash_i,i∈[1,D]The Hash fingerprint can uniquely represent the img of the medical CT image_i,i∈[1,D]。

Step 3: hash 64-bit fingerprint hash_i,i∈[1,D]And establishing a multi-bit index for storage.

Step 4: and generating a 64-bit hash fingerprint rethan for the medical CT image to be retrieved by the restig according to the process of generating the hash fingerprint in Step2.

Step 5: searching the hash fingerprint hash similar to the hash fingerprint rethrash of the medical CT image rettimg to be searched in the database according to a multi-bit search algorithm_i,i∈[1,D]If the Hamming distance N of the two Hash fingerprints is less than or equal to alpha, the Hash fingerprints are hashed_i,i∈[1,D]Corresponding medical CT image img_i,i∈[1,D]Label attribute label_i,i∈[1,D]For example, α is usually set to 3.

Further, in Step1, the medical CT images include a normal medical CT image and a patient medical CT image, and the number of the medical CT images is uniformly distributed.

Further, in the Step2, the medical CT image img_i,i∈[1,D]The generated hash fingerprint hash_i,i∈[1,D]The specific implementation steps are as follows:

step2.1: img medical CT image_i,i∈[1,D]Is compressed to 32 x 32, or cut into small squares of size 32 x 32.

Step2.2: img the compressed medical CT image_i,i∈[1,D]Converted into 256-step gray scale image and expressed as matrix I_i,i∈[1,D]In which I_i(m,k),i∈[1,D],m∈[1,32],k∈[1,32]Is the matrix element value.

Step2.3: calculating the matrix I_i,i∈[1,D]DCT matrix T_i,i∈[1,D]Where the matrix element values are denoted T_i(m,k),i∈[1,D],m∈[1,32],k∈[1,32]。

Step2.4: preserving DCT matrix T_i,i∈[1,D]The 8 x 8 low frequency part of the upper left corner of (1) and the rest of the part is deleted.

Step2.5: computing a DCT matrix T_i,i∈[1,D]Average value u of_i,i∈[1,D]Will DCT matrix T_i,i∈[1,D]According to the rule that m is 1 → 8, and k is 1 → 8 from left to right from top to bottom, the medical CT image img is generated by performing binarization according to the formula (1) and connecting the binarizations_i,i∈[1,D]64-bit hash fingerprint hash_i,i∈[1,D]。

Further, in Step3, the multi-bit index refers to hash of the fingerprint_i,i∈[1,D]Segmentation is performed according to M segments, as shown in formula (2), where M > α, hash_i,j,i∈[1,D],j∈[1,M]Representing a hash_i,i∈[1,D]Then hash the fingerprint, and_i,i∈[1,D]m segment sub-fingerprint hash_i,j,i∈[1,D],j∈[1,M]Performing (M-alpha) bit permutation and combination to establish an inverted index table, namely, hash of each hash fingerprint_i,i∈[1,D]Will haveAn index points to it and is marked as index 1, index 2, … … and index from top to bottom

The inverted index table when M is 5 is shown in formula (3); all hash fingerprint hashes in database_i,i∈[1,D]The inverted index table of (1) is called an inverted index total library.

hash_i＝[hash_i,1,hash_i,2,…,hash_i,M],i∈[1,D](2)

Further, in Step5, the implementation steps of the multi-bit search algorithm are as follows:

step5.1: performing M-segment segmentation on the Hash fingerprint rethrash of the medical CT image to be retrieved, as shown in formula (4), and performing (M-alpha) bit arrangement and combination according to the method described by Step3 to establish an inverted index table; the inverted index table when M is 5 is shown in formula (5);

rethash＝[rethash₁,rethash₂,…,rethash_M](4)

step5.2: searching a hash fingerprint set [ hash ] with the same index number and index value as the hash fingerprint rethan of the medical CT image retemg to be searched in the reverse index total library₁,hash₂,…,hash_K]And K represents the number of the hash fingerprints with the same index number and index value, and is called as a hash fingerprint set to be compared.

Step5.3: calculating Hash fingerprint rethan of medical CT image retetg to be retrieved and Hash fingerprint set [ hash ] to be compared₁,hash₂,…,hash_K]Every hash fingerprint hash_k,k∈[1,K]The Hamming distance N, if N is less than or equal to alpha, the hash fingerprint is hash_k,k∈[1,K]Corresponding medical CT image img_i,i∈[1,D]Label attribute label_i,i∈[1,D]And the like.

The invention has the beneficial effects that: compared with the prior art, the method mainly solves the problems of low efficiency, disease boundary and the like existing in the prior art when the similarity search of massive medical CT images is carried out, optimizes the search efficiency while keeping the similarity precision, and realizes the cross-disease similarity search.

Drawings

FIG. 1 is a flow chart of the steps of the present invention;

fig. 2 is an inverted index representation intent established when M is 5 in an embodiment of the present invention;

fig. 3 is an inverted index representation intent established when M is 6 in the embodiment of the present invention.

Detailed Description

The invention is further described with reference to the following drawings and detailed description.