Non-small cell lung cancer curative effect prediction method, system and device and storage medium

Abstract

The invention discloses a non-small cell lung cancer curative effect prediction method, system and device and a storage medium, and belongs to the technical field of medical image processing. The method comprises the following steps: acquiring a non-small cell lung cancer three-dimensional CT image; extracting a local prior dependency relationship between the image blocks by using 3D convolution; converting the local prior dependency relationship into local features; mining global features by using a self-attention module, and fusing the global features and local features by using a local cross attention module; and decoding to respectively obtain global, local and comprehensive characteristic curative effects of the non-small cell lung cancer. According to the method, efficient extraction of local features and feature fusion of global context are realized by combining a local feature extraction network and a feature conversion mechanism; furthermore, through multi-scale feature extraction and a cross attention mechanism, the understanding ability and the cross-patient generalization ability of a focus area are greatly improved, and the curative effect of the non-small cell lung cancer can be accurately, reliably and intelligently evaluated.

Description

Technical Field

[0001] The present invention relates to a method, system, device and storage medium for predicting the efficacy of non-small cell lung cancer, and belongs to the technical field of medical image processing. Background Art

[0002] Non-small cell lung cancer (NSCLC) is one of the main causes of cancer-related deaths globally. At present, the clinical evaluation of the efficacy of NSCLC patients mainly relies on imaging examinations and doctors' subjective judgments. However, these methods are highly subjective and are easily affected by doctors' experience levels, resulting in inconsistent evaluation results. In addition, traditional imaging analysis methods usually lack the ability to deeply mine global and local features, and cannot fully utilize the rich information in the image data, limiting the accuracy and reliability of the efficacy evaluation.

[0003] In recent years, significant progress has been made in medical image analysis technologies based on deep learning, especially in tasks such as tumor detection, segmentation, and staging. However, these methods still face the following challenges in practical applications: 1. Insufficient extraction and utilization of three-dimensional information: A large amount of medical image data is stored in three-dimensional form, and traditional two-dimensional convolutional networks are difficult to fully capture the spatial information in the images, resulting in insufficient modeling ability for the global structure of tumors. 2. Poor multi-scale feature fusion: There may be significant differences in the size, shape, and density of cancer lesion areas. Existing models have certain limitations in the processing and fusion of multi-scale features. 3. Insufficient interaction between global and local information: Traditional convolutional neural networks usually lack a mechanism for effectively interacting global and local features, limiting the model's in-depth mining of the internal features of tumor lesions. 4. Poor generality and adaptability of intelligent evaluation: There are significant individual differences in the imaging manifestations of tumors in different patients, and existing algorithms are difficult to meet the generalization requirements across patients and datasets.

[0004] Therefore, the current medical image analysis technologies based on deep learning have problems such as insufficient modeling ability and lack of feature extraction ability, and cannot comprehensively and accurately deeply mine image features, affecting the accuracy of diagnoses such as tumor detection and efficacy prediction, and thus unable to provide personalized and targeted precise medical services for patients. Summary of the Invention

[0005] In order to improve the accuracy of predicting the efficacy of non-small cell lung cancer, the present invention provides a method, system, device and storage medium for predicting the efficacy of non-small cell lung cancer, and the technical solutions are as follows:

[0006] The present invention provides a method for predicting the efficacy of non-small cell lung cancer, including:

[0007] Step 1: Obtain three-dimensional CT images of non-small cell lung cancer;

[0008] Step 2: Use 3D convolution to extract the local prior dependencies F between the three-dimensional CT image patches of non-small cell lung cancer local ;

[0009] Step 3: Convert the local prior dependencies into local features of the three-dimensional CT image patches of non-small cell lung cancer. The conversion process is as follows:

[0010] Perform a Flatten operation on the local prior dependencies F local to obtain a two-dimensional feature matrix:

[0011] F flat = Flatten(F local ), F flat ∈R N×C′

[0012] where N is the number of flattened features and C' is the dimension of each feature;

[0013] Initialize the learnable token matrix Q ∈ R M×C′ , where M represents the number of tokens. Use the token matrix Q as the query Q, and F flat as the key K and value V respectively. Calculate the local feature tokens through the multi-head attention mechanism. The attention calculation formula is:

[0014]

[0015] Map the attention output through a feed-forward network to obtain the local feature tokens

[0016]

[0017] where d s is the global feature dimension;

[0018] Step 4: Use the self-attention module to mine the global features of the three-dimensional CT image patches of non-small cell lung cancer, and use the local cross-attention module to interact the global features with the local features to obtain the fused features;

[0019] Step 5: Decode the global features, local features, and fusion to obtain the global feature efficacy, local feature efficacy, and comprehensive feature efficacy of non-small cell lung cancer respectively.

[0020] Optionally, the process of calculating the fused features in Step 4 includes:

[0021] Perform block embedding on the input data X:

[0022]

[0023] Where P is the number of blocks;

[0024] In the first six layers of the network, the standard window attention mechanism is used to process F patch as follows:

[0025]

[0026] Where W is the window partitioning operation;

[0027] In the last six layers of the network, each layer includes two steps:

[0028] First, extract global features through the window attention mechanism:

[0029]

[0030] Second, use the cross-attention mechanism to promote the fusion of local features and global features:

[0031]

[0032] The above formula represents the interaction between the local token feature T generated by the local feature conversion module qf and the global feature.

[0033] The present invention provides a non-small cell lung cancer treatment efficacy prediction system, and the system includes:

[0034] An image acquisition module configured to acquire an input three-dimensional CT image of non-small cell lung cancer;

[0035] A local feature encoder configured to extract local prior dependencies F between three-dimensional CT image blocks of non-small cell lung cancer using 3D convolution local ;

[0036] A local feature conversion module configured to convert the local prior dependencies into local features of three-dimensional CT image blocks of non-small cell lung cancer, and the conversion process is as follows:

[0037] Perform a Flatten operation on the local prior dependency F local to obtain a two-dimensional feature matrix:

[0038] F flat = Flatten(F local ), F flat ∈ R N×C′

[0039] Where N is the number of flattened features and C' is the dimension of each feature;

[0040] Initialize the learnable token matrix Q ∈ R M×C′, where M represents the number of tokens, and the token matrix Q is used as the query Q, and F flat are used as the key K and the value V respectively, and the local feature tokens are calculated through the multi-head attention mechanism. The attention calculation formula is:

[0041]

[0042] The attention output is mapped through a feed-forward network to obtain local feature tokens

[0043]

[0044] where d s is the global feature dimension;

[0045] The global feature encoder is configured to utilize the self-attention module to mine the global features of the three-dimensional CT image patches of non-small cell lung cancer, and utilize the local cross-attention module to interact the global features with the local features to obtain fused features;

[0046] The efficacy task feature decoder includes a local feature task decoder, a global feature task decoder, and a comprehensive feature task decoder. The local feature task decoder is used to decode the efficacy of the local features of non-small cell lung cancer, the global feature task decoder is used to decode the efficacy of the global features of non-small cell lung cancer, and the comprehensive feature task decoder is used to decode the efficacy of the comprehensive features of non-small cell lung cancer.

[0047] Optionally, the process of the global feature encoder calculating the fused features includes:

[0048] Performing block embedding on the input data X:

[0049]

[0050] where P is the number of blocks;

[0051] In the first 6 layers of the network, the standard window attention mechanism is used to process F patch as follows:

[0052]

[0053] where W is the window partitioning operation;

[0054] In the last 6 layers of the network, each layer includes two steps of operations:

[0055] The first step is to extract global features through the window attention mechanism:

[0056]

[0057] Step 2: Use the cross-attention mechanism to promote the fusion of local features and global features:

[0058]

[0059] The above formula represents the use of the local token feature T generated by the local feature transformation module qf to interact with the global feature.

[0060] Optionally, the local feature encoder consists of a series of 3D convolutional layers, and each 3D convolutional layer is followed by a batch normalization layer, a ReLU activation function, and a 3D max pooling layer. At the same time, the learning ability of the local feature encoder is enhanced through residual connections.

[0061] Optionally, in the efficacy task feature decoder, each decoder consists of two fully connected layers and a ReLU activation function.

[0062] Optionally, the loss function for model training is:

[0063]

[0064] where represents the local feature loss, represents the global feature loss, represents the fused feature loss, and α and β respectively represent the weights of the local feature loss and the global feature loss.

[0065] Optionally, the loss function for model training adopts the binary cross-entropy loss function.

[0066] The present invention provides a non-small cell lung cancer efficacy prediction device, including a memory and a processor;

[0067] The memory is used to store a computer program;

[0068] The processor is used to implement the non-small cell lung cancer efficacy prediction method as described in any one of the above when executing the computer program.

[0069] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the non-small cell lung cancer efficacy prediction method as described in any one of the above is implemented.

[0070] The beneficial effects of the present invention are:

[0071] The efficacy prediction model constructed in the present invention efficiently combines a 3D convolutional neural network and a self-attention mechanism. The 3D convolutional neural network is used to capture the three-dimensional spatial features of the image data, and the self-attention mechanism is utilized to introduce global context information to significantly improve the comprehensiveness and accuracy of feature extraction. Compared with the prior art, the present invention realizes the efficient extraction of local features and the feature fusion of global context by combining a local feature extraction network and a feature transformation mechanism; by further integrating the multi-scale feature extraction ability and cross-attention mechanism of the Transformer, the understanding ability of the lesion area and the generalization ability across patients are greatly improved. The experimental results prove that the prediction model constructed in the present invention shows good performance in terms of classification and accuracy, thus enabling accurate and reliable intelligent evaluation of the efficacy of non-small cell lung cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0073] Figure 1 It is a schematic diagram of the basic process of the non-small cell lung cancer efficacy prediction method of the present invention.

[0074] Figure 2 It is a schematic diagram of the structure of the local feature encoder provided in the second embodiment of the present invention.

[0075] Figure 3 It is a schematic diagram of the structure of the local feature transformation module provided in the second embodiment of the present invention.

[0076] Figure 4 It is a schematic diagram of the structure of the global feature encoder provided in the second embodiment of the present invention.

[0077] Figure 5 It is a schematic diagram of the structure of the fusion of global features and local features provided in the second embodiment of the present invention.

[0078] Figure 6 It is an AUROC index effect diagram of the non-small cell lung cancer efficacy prediction model provided in the third embodiment of the present invention.

[0079] Figure 7 It is an AUPRC index effect diagram of the non-small cell lung cancer efficacy prediction model provided in the third embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0080] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the drawings.

[0081] Example 1:

[0082] This example provides a method for predicting the efficacy of non-small cell lung cancer, including:

[0083] Step 1: Obtain three-dimensional CT images of non-small cell lung cancer;

[0084] Step 2: Use 3D convolution to extract the local prior dependence relationship F between three-dimensional CT image blocks of non-small cell lung cancer local ;

[0085] Step 3: Convert the local prior dependence relationship into local features of three-dimensional CT image blocks of non-small cell lung cancer. The conversion process is as follows:

[0086] Perform a Flatten operation on the local prior dependence relationship F local to obtain a two-dimensional feature matrix:

[0087] F flat = Flatten(F local ), F flat ∈R N×C′

[0088] where N is the number of flattened features and C' is the dimension of each feature;

[0089] Initialize a learnable token matrix Q ∈ R M×C′ , where M represents the number of tokens. Use the token matrix Q as the query Q, and F flat as the key K and value V respectively. Calculate the local feature tokens through the multi-head attention mechanism. The attention calculation formula is:

[0090]

[0091] Map the attention output through a feed-forward network to obtain local feature tokens

[0092]

[0093] where d s is the global feature dimension;

[0094] Step 4: Use the self-attention module to mine the global features of three-dimensional CT image blocks of non-small cell lung cancer, and use the local cross-attention module to interact the global features with the local features to obtain fused features;

[0095] Perform block embedding on the input data X:

[0096]

[0097] Among them, P is the number of chunks;

[0098] In the first six layers of the network, the standard window attention mechanism is used to process F patch as follows:

[0099]

[0100] Among them, W is the window partitioning operation;

[0101] In the last six layers of the network, each layer includes two steps:

[0102] The first step is to extract global features through the window attention mechanism:

[0103]

[0104] The second step is to use the cross-attention mechanism to promote the fusion of local features and global features:

[0105]

[0106] The above formula represents the interaction between the local token feature T generated by the local feature conversion module qf and the global feature.

[0107] Step 5: Decode the global feature, local feature, and fusion to obtain the global feature efficacy, local feature efficacy, and comprehensive feature efficacy of non-small cell lung cancer respectively.

[0108] Example 2:

[0109] This example provides a non-small cell lung cancer efficacy prediction model based on a 3D convolutional neural network, including: an image acquisition module, a local feature encoder, a local feature conversion module, a global feature encoder, and an efficacy task feature decoder.

[0110] Among them, the image acquisition module is used to acquire the input three-dimensional CT image of non-small cell lung cancer.

[0111] The local feature encoder is used to extract the local prior dependence relationship between the three-dimensional CT image blocks of non-small cell lung cancer, as Figure 2 shown. This process is implemented through a series of 3D convolutional layers, and each 3D convolutional layer is followed by a batch normalization layer, a ReLU activation function, and a 3D max pooling layer. At the same time, the learning ability of the local feature encoder is enhanced through residual connections, enabling it to effectively extract local features.

[0112] The local feature conversion module is used to convert the local prior dependence relationship between the three-dimensional CT image blocks of non-small cell lung cancer into the local features of the three-dimensional CT image blocks of non-small cell lung cancer, as Figure 3As shown, the transformation layer includes the following steps:

[0113] Initialize a learnable token Q of fixed length in R M×C′ , where M represents the number of tokens.

[0114] Perform a Flatten operation on the features F output by the local feature encoder local to obtain a two-dimensional feature matrix:

[0115] F flat = Flatten(F local ), F flat ∈R N×C′

[0116] where N = T'×H'×W' is the number of flattened features, and C' is the dimension of each feature.

[0117] Initialize a learnable token matrix Q ∈ R M×C′ , where M represents the number of tokens, take Q as the query Q, and F flat as the key K and value V respectively, and calculate the local feature tokens through the multi-head attention mechanism. The attention calculation formula is as follows:

[0118]

[0119] Map the attention output through a feed-forward network (FFN) to obtain the local feature tokens

[0120]

[0121] where d s is the global feature dimension.

[0122] The global feature encoder includes: a self-attention module and a local cross-attention module. The self-attention module is used to mine the global features of the three-dimensional CT image patches of non-small cell lung cancer, and the local cross-attention module is used to interact the global features with the local features of the three-dimensional CT image patches of non-small cell lung cancer. The specific processing process is as follows:

[0123] Perform block embedding on the input data X:

[0124]

[0125] where P is the number of blocks.

[0126] As Figure 5 shown, in the first 6 layers of the network, use the standard window attention mechanism to process F patch :

[0127]

[0128] Among them, W is the window partitioning operation.

[0129] As Figure 6 shown, in the latter 6 layers of the network, each layer includes two steps of operations:

[0130] In the first step, global features are extracted through the window attention mechanism:

[0131]

[0132] In the second step, the cross-attention mechanism is used to promote the fusion of local features and global features: Using the local token feature T generated by the local feature conversion module qf to interact with the global features:

[0133]

[0134] The efficacy task feature decoder includes: a local feature task decoder, a global feature task decoder, and a comprehensive feature task decoder. The local feature task decoder is used to decode the efficacy of local features of non-small cell lung cancer, the global feature task decoder is used to decode the efficacy of global features of non-small cell lung cancer, and the comprehensive feature task decoder is used to decode the efficacy of comprehensive features of non-small cell lung cancer. Each decoder consists of two fully connected layers and a ReLU activation function.

[0135] During the decoding process, the local feature loss is represented by the following binary cross-entropy formula:

[0136]

[0137] Among them, p local,i = σ(Wlocal·F local,i + b local ).

[0138] Similarly, the global and comprehensive feature losses are respectively:

[0139]

[0140] The final total loss function is the weighted sum of the losses of each part:

[0141]

[0142] Among them, α, β are weight coefficients used to balance the importance of the losses of each part.

[0143] Example 3:

[0144] To verify the effectiveness of the method of the present invention, in this embodiment, three-dimensional CT images of non-small cell lung cancer are used to train and test a non-small cell lung cancer treatment efficacy prediction model.

[0145] The image data used in this embodiment comes from real non-small cell lung cancer patients. All CT images are collected and processed in a standardized format. The image size is 256×256×64, and image sampling is performed at a 10-fold magnification. Each sample corresponds to a three-dimensional CT image, and it is marked whether the patient has achieved pathological complete remission (pCR).

[0146] First, the data set is divided into a training set and a test set. The training set contains 400 cases (about 80%), and the test set contains 100 cases (about 20%). To ensure the representativeness of the data, the cases in the training set and the test set are both randomly sampled, and the quality inspection and verification are performed on the CT images corresponding to each sample after annotation.

[0147] In the model training stage, this embodiment adopts the following steps:

[0148] Data preprocessing: For each three-dimensional CT image, data augmentation techniques (such as rotation, scaling, etc.) are used for processing to expand the number of training samples. Then, the pixel values of all images are normalized to the interval [0,1] for input into the neural network for training.

[0149] The model architecture includes: a local feature encoder, a local feature transformation module, a global feature encoder, and a treatment efficacy task feature decoder. The local feature encoder uses multiple 3×3×3 convolutional layers to extract local features, and uses the ReLU activation function and max pooling operation for non-linear mapping. The global feature encoder combines the self-attention mechanism and the local cross-attention module to extract global information and fuse it with local features, and finally performs treatment efficacy prediction through the treatment efficacy task feature decoder.

[0150] Training process: The model is trained using the Adam optimizer and optimized using a multi-task loss function. The loss function includes local feature loss, global feature loss, and comprehensive feature loss. During the training process, this embodiment sets the learning rate to 0.001 and performs 4 complete trainings with a training cycle of 50 rounds.

[0151] Evaluation metrics: After each round of training, the evaluation is performed on the test set, and two metrics, AUROC (area under the receiver operating characteristic curve) and AUPRC (area under the precision-recall curve), are used to evaluate the prediction effect of the model. The AUROC and AUPRC results of the model are as Figure 6 and Figure 7 shown, respectively demonstrating the performance of the model at different training stages.

[0152] AUROC curve: It shows the performance of the model's Receiver Operating Characteristic curve (ROC curve) on the test set, reflecting the model's ability to distinguish different classes.

[0153] AUPRC curve: It shows the performance of the model's Precision-Recall curve (PR curve) on the test set, evaluating the model's effect when dealing with imbalanced data.

[0154] Through 4 independent trainings and tests, the model of this embodiment shows good performance in both AUROC and AUPRC metrics.

[0155] Some steps in the embodiments of the present invention can be implemented by software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk, etc.

[0156] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the efficacy of non-small cell lung cancer, characterized in that, The method includes: Step 1: Obtain three-dimensional CT images of non-small cell lung cancer; Step 2: Extract the local prior dependencies F between the 3D CT image patches of non-small cell lung cancer using 3D convolution local ; Step 3: Convert the local prior dependency into local features of three-dimensional CT image patches of non-small cell lung cancer. The conversion process is as follows: Perform a Flatten operation on the local prior dependency relationship F local to obtain a two-dimensional feature matrix: F flat = Flatten(F local ), F flat ∈R N×C′ Where N is the number of flattened features, and C' is the dimension of each feature; Initialize the learnable token matrix Q ∈ R M×C′ , where M represents the number of tokens. Use the token matrix Q as the query Q, and F flat as the key K and value V respectively. Calculate the local feature tokens through the multi-head attention mechanism. The attention calculation formula is: Map the attention output through a feed-forward network to obtain local features where d s is the global feature dimension; Step 4: Use the self-attention module to mine the global features of three-dimensional CT image patches of non-small cell lung cancer, and use the local cross-attention module to interact the global features with the local features to obtain fused features; Step 5: Decode the global features, local features, and fusion to obtain the global feature efficacy, local feature efficacy, and comprehensive feature efficacy of non-small cell lung cancer respectively.

2. The non-small cell lung cancer treatment efficacy prediction method according to claim 1, wherein, The process of calculating the fused features in Step 4 includes: Perform block embedding on the input data X: Where P is the number of blocks; In the first six-layer network, the standard window attention mechanism is used to process F patch for processing: Where W is the window partitioning operation; In the last 6 layers of the network, each layer includes two operations: First, extract global features through the window attention mechanism: Second, use the cross-attention mechanism to promote the fusion of local features and global features: The above formula represents the local token feature T generated by the local feature transformation module qf interacts with the global feature.

3. A non-small cell lung cancer treatment efficacy prediction system, characterized in that, The system includes: An image acquisition module configured to obtain input three-dimensional CT images of non-small cell lung cancer; A local feature encoder, configured to utilize 3D convolution to extract local prior dependencies F between 3D CT image patches of non-small cell lung cancer local ; A local feature conversion module configured to convert the local prior dependency into local features of three-dimensional CT image patches of non-small cell lung cancer. The conversion process is as follows: Perform a Flatten operation on the local prior dependency relationship F local to obtain a two-dimensional feature matrix: F flat = Flatten(F local ), F flat ∈ R N×C′ Where N is the number of flattened features, and C' is the dimension of each feature; Initialize the learnable token matrix Q ∈ R M×C′ , where M represents the number of tokens. Use the token matrix Q as the query Q, and F flat are used as the key K and value V respectively. Calculate the local feature tokens through the multi-head attention mechanism. The attention calculation formula is: Map the attention output through a feed-forward network to obtain local features Among them, d s is the global feature dimension; A global feature encoder configured to use the self-attention module to mine the global features of three-dimensional CT image patches of non-small cell lung cancer, and use the local cross-attention module to interact the global features with the local features to obtain fused features; A treatment efficacy task feature decoder includes a local feature task decoder, a global feature task decoder, and a comprehensive feature task decoder. The local feature task decoder is used to decode the local feature efficacy of non-small cell lung cancer, the global feature task decoder is used to decode the global feature efficacy of non-small cell lung cancer, and the comprehensive feature task decoder is used to decode the comprehensive feature efficacy of non-small cell lung cancer.

4. The non-small cell lung cancer treatment efficacy prediction system according to claim 3, characterized in that, The process of calculating the fused features by the global feature encoder includes: Perform block embedding on the input data X: Where P is the number of blocks; In the first six layers of the network, the standard window attention mechanism is used to process F patch as follows: Where W is the window partitioning operation; In the last 6 layers of the network, each layer includes two operations: First, extract global features through the window attention mechanism: Second, use the cross-attention mechanism to promote the fusion of local features and global features: The above formula represents the local token feature T generated by the local feature transformation module qf interacting with the global feature.

5. The non-small cell lung cancer treatment efficacy prediction system according to claim 3, characterized in that, The local feature encoder is composed of a series of 3D convolutional layers. Each 3D convolutional layer is followed by a batch normalization layer, a ReLU activation function, and a 3D max pooling layer. At the same time, the learning ability of the local feature encoder is enhanced through residual connections.

6. The non-small cell lung cancer treatment efficacy prediction system according to claim 3, wherein, In the treatment efficacy task feature decoder, each decoder consists of two fully connected layers and a ReLU activation function.

7. The non-small cell lung cancer treatment efficacy prediction system according to claim 3, wherein The loss function for model training is: Among them, represents the local feature loss, represents the global feature loss, represents the fused feature loss, and α and β respectively represent the weights of the local feature loss and the global feature loss.

8. The non-small cell lung cancer treatment efficacy prediction system according to claim 7, characterized in that, The loss function for model training uses the binary cross-entropy loss function.

9. A non-small cell lung cancer treatment efficacy prediction device, characterized in that, Includes a memory and a processor; The memory is used to store computer programs; The processor is used to implement the non-small cell lung cancer treatment efficacy prediction method as described in any one of claims 1 to 2 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by a processor, the non-small cell lung cancer treatment efficacy prediction method according to any one of claims 1 to 2 is implemented.

Images