Medical image segmentation method and device for pheochromocytoma, equipment and medium

Abstract

Embodiments of the present application provide a pheochromocytoma medical image segmentation method and device, equipment and medium. Among them, the method comprises: acquiring a pheochromocytoma medical image; inputting the pheochromocytoma medical image into a pheochromocytoma segmentation model for image segmentation to obtain a segmentation result of the pheochromocytoma medical image, the pheochromocytoma segmentation model is composed of a semi-supervised segmentation model and a SAM-Med3D segmentation model, the semi-supervised segmentation model includes a shared encoder and multiple independent decoders, a multi-scale attention module is arranged on the encoding path of the encoder, the multi-scale attention module is used to strengthen the extraction of key features of pheochromocytoma, and the pseudo label is selected by competition and comparison among the outputs of the multiple decoders, and the position information generated by the multiple decoders is used as the source of mask prompt information of the SAM-Med3D segmentation model. Based on this, the embodiments of the present application can improve the automatic segmentation performance of the pheochromocytoma medical image.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing technology, and in particular to a method, apparatus, device and medium for medical image segmentation of pheochromocytoma. Background Technology

[0002] Pheochromocytoma is a rare neuroendocrine tumor with an annual incidence of approximately 0.8 per 100,000 people. Accurate diagnosis and timely surgical intervention can prevent tumor progression and metastasis, reducing mortality. However, the diverse clinical presentation of pheochromocytoma poses challenges to accurate diagnosis. Furthermore, pheochromocytomas are typically located in the abdomen, surrounded by important multi-organ anatomical structures. As the tumor grows and becomes malignant, it can invade adjacent organs, further increasing surgical risks. Therefore, accurately identifying the tumor location and precisely delineating its boundaries to facilitate surgical treatment is crucial.

[0003] Existing pheochromocytoma segmentation methods require sufficient labels to ensure model segmentation performance. For 3D medical images, obtaining pixel-level annotations is extremely time-consuming and difficult, as accurate annotations can only be provided by experts with domain knowledge. For pheochromocytoma segmentation tasks, annotating CT volumes can take several hours. The complex shape and appearance of the tumor further increase the workload and time required for annotation, making it difficult to annotate a large number of 3D pheochromocytoma CT scans for training. Furthermore, the inherent properties of pheochromocytoma also pose challenges to automated segmentation. First, tumor size varies greatly depending on disease severity. A small tumor lesion may contain only a few voxels, while a large tumor can occupy a large space, easily leading to class imbalance. Second, the tumor location is not fixed, appearing in different locations in the abdomen with a complex spatial distribution. In addition, the tumor morphology varies considerably; smaller tumors are usually homogeneous, while larger tumors may have uneven grayscale, exhibiting central necrosis, which may lead to undersegmentation. Moreover, the tumor's proximity to abdominal organs may cause the model to misinterpret the morphological changes of these organs as tumor boundaries, incorrectly segmenting them as part of the tumor, thus causing oversegmentation. These factors, coupled with the low contrast of soft tissue in abdominal CT images, further exacerbate the difficulty of pheochromocytoma segmentation, resulting in insufficient performance in automated segmentation. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for pheochromocytoma medical image segmentation. By applying semi-supervised techniques to pheochromocytoma segmentation, the problem of insufficient labeled data is effectively alleviated. Furthermore, pseudo-labels are selected from multiple decoder outputs, which helps improve the reliability of pseudo-labels and introduces uncertainty assessment to reduce the impact of pseudo-label noise. A multi-scale attention module is employed to enhance the extraction of key features of pheochromocytoma, effectively addressing challenges such as variations in size and location of pheochromocytoma and low contrast. The pseudo-labels generated by the semi-supervised segmentation model are used as a source of prompting information for SAM-Med3D, effectively solving the shortcomings of labor-intensive manual prompting, thereby further improving the automated segmentation performance of pheochromocytoma medical images.

[0005] In a first aspect, embodiments of the present invention provide a medical image segmentation method for pheochromocytoma, comprising:

[0006] Acquiring medical images of pheochromocytoma;

[0007] The pheochromocytoma medical image is input into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder. The multi-scale attention module is used to enhance the extraction of key features of pheochromocytoma. False labels are selected by competitive comparison among the outputs of multiple decoders, and the position information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model.

[0008] In some embodiments, the step of inputting the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image includes:

[0009] The pheochromocytoma medical image is input into the encoder for feature extraction to obtain the pheochromocytoma medical image features.

[0010] The medical image features of the pheochromocytoma are input into multiple decoders to obtain multiple output results;

[0011] A dynamic uncertainty correction strategy is used to select pseudo-labels from multiple output results;

[0012] The pseudo-labels are filtered to generate masked prompt information;

[0013] The SAM-Med3D segmentation model is guided by the mask prompt information to perform image segmentation on the pheochromocytoma medical image, thereby obtaining the segmentation result of the pheochromocytoma medical image.

[0014] In some embodiments, the multi-scale attention module includes an MSA-H variant applied at high resolution and an MSA-L variant applied at low resolution. The step of inputting the pheochromocytoma medical image into the encoder for feature extraction to obtain pheochromocytoma medical image features includes:

[0015] The pheochromocytoma medical image is input into the encoder to learn variable pheochromocytoma features;

[0016] The variable pheochromocytoma features are input into the multi-scale attention module to extract multi-scale medical image features of the pheochromocytoma.

[0017] In some embodiments, the step of selecting pseudo-labels from multiple output results using a dynamic uncertainty correction strategy includes:

[0018] The optimal prediction result is selected as the pseudo-label from multiple output results through dynamic competition.

[0019] An uncertainty correction method is introduced to evaluate the selected pseudo-labels at the voxel level to obtain the confidence level of each voxel;

[0020] Assign a weight factor value to each voxel based on its confidence level.

[0021] In some embodiments, assigning an uncertainty weighting factor to each voxel based on its confidence level includes:

[0022] Increase the weight factor value of the voxels with high confidence;

[0023] Reduce the weight factor value of the voxels with low confidence.

[0024] In some embodiments, the method further includes:

[0025] The output of the optimal decoder is selected from the multiple decoder outputs and used as a pseudo-label to guide the suboptimal decoder.

[0026] In some embodiments, the training method for the pheochromocytoma segmentation model includes:

[0027] Determine the final loss function, which is determined based on multiple supervised loss functions, multiple uncertain loss functions, and multiple balancing hyperparameters;

[0028] The pheochromocytoma segmentation model is trained based on the final loss function to obtain the trained pheochromocytoma segmentation model.

[0029] Secondly, embodiments of the present invention also provide a medical image segmentation device for pheochromocytoma, the device comprising:

[0030] The acquisition module is used to acquire medical images of pheochromocytoma;

[0031] The segmentation module is used to input the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation, and obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder. The multi-scale attention module is used to enhance the extraction of key features of pheochromocytoma. False labels are selected by competitive comparison among the outputs of multiple decoders, and the position information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model.

[0032] Thirdly, embodiments of the present invention also provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the pheochromocytoma medical image segmentation method as described in the first aspect.

[0033] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions for performing the pheochromocytoma medical image segmentation method as described in the first aspect.

[0034] The present invention provides a method, apparatus, device, and medium for pheochromocytoma medical image segmentation. The pheochromocytoma medical image segmentation method includes: acquiring a pheochromocytoma medical image; inputting the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain a segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder to enhance the extraction of key features of pheochromocytoma. False labels are selected through competitive comparison among the outputs of multiple decoders, and the positional information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model. This invention effectively alleviates the problem of insufficient label data by applying semi-supervised technology to pheochromocytoma segmentation. Furthermore, selecting false labels from the outputs of multiple decoders helps improve the reliability of false labels and introduces uncertainty assessment to reduce the impact of false label noise. Furthermore, a multi-scale attention module is employed to enhance the extraction of key features of pheochromocytoma, effectively addressing challenges such as variations in size and location of pheochromocytoma and low contrast. Pseudo-labels generated by a semi-supervised segmentation model are used as the source of cue information for SAM-Med3D, effectively overcoming the shortcomings of labor-intensive manual cueing and further improving the automated segmentation performance of pheochromocytoma medical images. Attached Figure Description

[0035] Figure 1 This is a flowchart of a medical image segmentation method for pheochromocytoma provided in one embodiment of the present invention;

[0036] Figure 2 This is a schematic diagram of the structure of a pheochromocytoma segmentation model provided in one embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the structure of a multi-scale attention module provided in one embodiment of the present invention;

[0038] Figure 4 This is a diagram of the pseudo-label selection process algorithm provided in one embodiment of the present invention;

[0039] Figure 5 This is a schematic diagram of the structure of a pheochromocytoma medical image segmentation device provided in one embodiment of the present invention;

[0040] Figure 6 This is a schematic diagram of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the following drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0043] In this embodiment of the invention, the terms "furthermore," "exemplarily," or "optionally" are used as examples, illustrations, or descriptions and should not be construed as being more preferred or advantageous than other embodiments or designs. The use of the terms "furthermore," "exemplarily," or "optionally" is intended to present the relevant concepts in a specific manner.

[0044] To facilitate a more convenient description of the working principle of the embodiments of the present invention, the following introduction of relevant technical scenarios is given first.

[0045] Pheochromocytoma is a rare neuroendocrine tumor with an annual incidence of approximately 0.8 per 100,000 people. Accurate diagnosis and timely surgical intervention can prevent tumor progression and metastasis, reducing mortality. However, the diverse clinical presentation of pheochromocytoma poses challenges to accurate diagnosis. Furthermore, pheochromocytomas are typically located in the abdomen, surrounded by important multi-organ anatomical structures. As the tumor grows and becomes malignant, it can invade adjacent organs, further increasing surgical risks. Therefore, accurately identifying the tumor location and precisely delineating its boundaries to facilitate surgical treatment is crucial.

[0046] Existing pheochromocytoma segmentation methods require sufficient labels to ensure model segmentation performance. For 3D medical images, obtaining pixel-level annotations is extremely time-consuming and difficult, as accurate annotations can only be provided by experts with domain knowledge. For pheochromocytoma segmentation tasks, annotating CT volumes can take several hours. The complex shape and appearance of the tumor further increase the workload and time required for annotation, making it difficult to annotate a large number of 3D pheochromocytoma CT scans for training. Furthermore, the inherent properties of pheochromocytoma also pose challenges to automated segmentation. First, tumor size varies greatly depending on disease severity. A small tumor lesion may contain only a few voxels, while a large tumor can occupy a large space, easily leading to class imbalance. Second, the tumor location is not fixed, appearing in different locations in the abdomen with a complex spatial distribution. In addition, the tumor morphology varies considerably; smaller tumors are usually homogeneous, while larger tumors may have uneven grayscale, exhibiting central necrosis, which may lead to undersegmentation. Moreover, the tumor's proximity to abdominal organs may cause the model to misinterpret the morphological changes of these organs as tumor boundaries, incorrectly segmenting them as part of the tumor, thus causing oversegmentation. These factors, coupled with the low contrast of soft tissue in abdominal CT images, further exacerbate the difficulty of pheochromocytoma segmentation, resulting in insufficient performance in automated segmentation.

[0047] Based on this, the present invention provides a method, apparatus, device, and medium for pheochromocytoma medical image segmentation. The pheochromocytoma medical image segmentation method includes: acquiring a pheochromocytoma medical image; inputting the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder to enhance the extraction of key features of pheochromocytoma. False labels are selected through competitive comparison among the outputs of multiple decoders, and the positional information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model. The present invention effectively alleviates the problem of insufficient labeled data by applying semi-supervised technology to pheochromocytoma segmentation. Furthermore, the selection of false labels from multiple decoder outputs helps improve the reliability of false labels, and the introduction of uncertainty assessment reduces the impact of false label noise. Furthermore, a multi-scale attention module is employed to enhance the extraction of key features of pheochromocytoma, effectively addressing challenges such as variations in size and location of pheochromocytoma and low contrast. Pseudo-labels generated by a semi-supervised segmentation model are used as the source of cue information for SAM-Med3D, effectively overcoming the shortcomings of labor-intensive manual cueing and further improving the automated segmentation performance of pheochromocytoma medical images.

[0048] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0049] like Figure 1 As shown, Figure 1 This is a flowchart of a medical image segmentation method for pheochromocytoma provided in an embodiment of the present invention. The medical image segmentation method for pheochromocytoma may include, but is not limited to, steps S101 to S102.

[0050] Step S101: Obtain medical images of pheochromocytoma;

[0051] Step S102: The pheochromocytoma medical image is input into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder to enhance the extraction of key features of pheochromocytoma. The pseudo-label is selected by competitive comparison among the outputs of multiple decoders, and the position information predicted by multiple decoders is used as the source of mask cue information for the SAM-Med3D segmentation model.

[0052] It is understandable that, such as Figure 2 As shown, the pheochromocytoma segmentation model of this invention mainly consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model is a dynamic uncertainty correction model, named DR-Net, which has a shared encoder and three independent decoders. To enhance key feature extraction, a multi-scale attention module is designed on the encoding path. For different decoder outputs, the prediction with the best performance is competitively selected as a pseudo-label to correct the model. DR-Net+ is formed by adding SAM-Med3D to DR-Net. SAM-Med3D serves as an additional supervisory branch to enhance the utilization of unlabeled data and assist in pheochromocytoma segmentation. Considering computational resource limitations, during DR-Net training, SAM-Med3D does not participate in backpropagation but directly performs inference based on the mask hints provided by DR-Net. The inference result acts as a supervisory term on DR-Net, thereby improving the segmentation performance of DR-Net.

[0053] Understandably, this invention proposes a novel semi-supervised framework for pheochromocytoma segmentation. This invention selects pseudo-labels from multiple decoder outputs, which helps improve the reliability of pseudo-labels and introduces uncertainty assessment to reduce the impact of pseudo-label noise. This invention proposes a multi-scale attention module to enhance the extraction of key features of pheochromocytoma, effectively addressing challenges such as variations in size and location of pheochromocytoma and low contrast. This invention utilizes pseudo-labels generated by the semi-supervised segmentation model as a source of cue information for SAM-Med3D, effectively solving the shortcomings of labor-intensive manual cueing.

[0054] Understandably, this invention is the first to apply semi-supervised techniques to pheochromocytoma segmentation, effectively alleviating the problem of insufficient labeled data. Existing pheochromocytoma segmentation methods require all labeled data to ensure segmentation performance. Furthermore, traditional level set-based methods also achieve good segmentation results, but are still limited by manual parameter tuning. The pheochromocytoma medical image segmentation method of this invention is an automated segmentation method that does not require manual parameter tuning. Moreover, this invention introduces prior knowledge from SAM-Med3D to further improve segmentation performance.

[0055] It is understood that, for ease of description of the method of this invention, the problem of the SSL task is described as follows, with n labeled data represented as follows: y i yes The labels are m unlabeled data points, represented as follows: Where n << m. The objective of this invention is to improve the performance of the pheochromocytoma segmentation framework using a small amount of labeled data and a large amount of unlabeled data.

[0056] Understandably, in the field of medical image segmentation, convolutional neural networks incorporating attention mechanisms have demonstrated strong advantages in challenging scenarios such as low contrast, irregular shapes, and small targets. However, in the pheochromocytoma segmentation task, when faced with challenges such as low contrast, shape and size variations, and especially when the pheochromocytoma is deformed irregularly due to compression from surrounding abdominal organs, the model struggles to make the most advantageous decision based on the target's contextual consistency. Furthermore, the close connection to surrounding abdominal organs can also lead to oversegmentation and undersegmentation issues.

[0057] To effectively address the aforementioned problems, this invention designs a multi-scale attention module (MSA) in the model encoding stage to enhance key feature extraction. The MSA appears at the beginning of each stage of the encoder, receiving the output (or input image) of the previous stage as input, learning variable pheochromocytoma features, and simultaneously extracting multi-scale tumor features. For example... Figure 3As shown, the multi-scale attention module has two variants: MSA-H for high resolution and MSA-L for low resolution. MSA-H and MSA-L are described below.

[0058] and Let W×H×D represent the feature input and output of MSA, respectively, and let C represent the spatial size. i and C o These represent the number of input and output channels, respectively. MSA-H first passes F through convolutional layers with kernel sizes of 1 and 3, respectively, followed by a batch normalization layer and a ReLU activation layer for each layer. Then, a downsampling layer reduces the feature resolution by half, followed by pointwise convolution to map the feature interactions between channels. Upsampling is then used to restore the feature resolution to obtain F″, which is then sent to a sigmoid activation function to obtain the attention coefficient α. This invention uses an average pooling layer with a stride and kernel size of 2 for downsampling, and trilinear interpolation for upsampling. Using downsampling increases the receptive field of the convolutional layers and reduces the computational cost of spatial attention. The obtained attention coefficient is multiplied by F, and different attention weights are assigned to all voxels in F, then added back to F to obtain the attention coefficient α. The expression is:

[0059]

[0060] At the bottleneck of the segmentation network, the feature map resolution is very low. Continuing to use the downsampling layers in MSA-H makes the feature map even smaller, to the point that spatial detail information cannot be preserved. Therefore, this invention proposes MSA-L, a variant of MSA-H. Due to the low spatial resolution, this invention removes the downsampling and upsampling operations, retaining only pointwise convolutions.

[0061] Understandably, generating pseudo-labels for unannotated images has proven effective for semi-supervised medical image segmentation. The key is improving the quality of the pseudo-labels; using incorrect pseudo-labels for correction can lead to a catastrophic decline in model performance. Selecting the best-performing pseudo-label from multiple options is considered an effective method, filtering out all predictions with suboptimal performance. This invention proposes a dynamic uncertainty correction strategy for processing unlabeled data. This strategy consists of two main parts: first, selecting the best-performing prediction from multiple different decoder outputs as the pseudo-label; and second, introducing an uncertainty assessment method to evaluate the selected pseudo-label at the voxel level, reducing the impact of high-uncertainty voxels.

[0062] Understandably, unlike selecting from multiple subnets, this invention dynamically selects pseudo-tags from the outputs of multiple independent decoders, which share the same encoder. This design has three advantages: first, it helps reduce representational differences introduced by different network parameters at the encoder stage; second, selecting from multiple outputs can effectively improve the reliability of pseudo-tags; and third, the multi-branch setup avoids the bias of a single prediction head and improves the diversity of pseudo-tags.

[0063] like Figure 4 As shown, Algorithm 1 describes the pseudo-label selection process. This invention uses Dice loss to detect the segmentation performance of multiple decoders in real time and selects the best-performing decoder. Dice loss is calculated on the labeled data in each batch. Among all the decoder outputs, the decoder with the best segmentation performance is generally considered to have higher reliability in predicting unlabeled data and can be regarded as a pseudo-label. It should be noted that although pseudo-labels have higher reliability in all decoder outputs, they still have a certain gap with the real labels.

[0064] It is understandable that the acquired pseudo-labels can be used to correct the suboptimal decoder output and improve its segmentation performance. This invention employs consistency regularization loss to achieve this correction:

[0065]

[0066] P l Indicates a pseudo tag. This represents the output of the k-th decoder. However, directly using the pseudo-labels generated by the optimal decoder for model correction is not the best choice because without real labels for guidance, the prediction results may be unreliable and noisy, which will affect model training. Therefore, this invention needs to evaluate the uncertainty of pseudo-labels, obtain the confidence level of each voxel, and then guide the model to learn from more reliable information. This invention first uses the KL divergence between the optimal decoder and the suboptimal decoder as the uncertainty:

[0067]

[0068] Let n represent the KL divergence. v Indicates the number of voxels. P represents the prediction of the k-th decoder output on the i-th voxel. i This represents the prediction of the i-th voxel in the pseudo-label. If the predictions of the optimal decoder and the k-th suboptimal decoder differ significantly, The value will be relatively large. Then... As an uncertainty weighting factor, the uncertainty loss function can be expressed as:

[0069]

[0070] By applying an uncertainty weighting factor, voxels with higher uncertainty are automatically assigned lower weights, while voxels with lower uncertainty automatically receive higher weights, thus enabling the model to learn more reliable information. This invention also addresses the uncertainty term... Introducing constraint terms can improve the robustness of the model. This can be achieved by reducing... The model effectively learns reliable representations from predictions to keep the decoder outputs close under different perturbations, thus improving the overall segmentation performance of the model.

[0071] Furthermore, based on the smoothness assumption of semi-supervised learning, this invention aims for the outputs of two independent suboptimal decoders to be similar. Therefore, this invention introduces another consistency regularization to further constrain the model, achieved through cosine distance:

[0072]

[0073] shrinking This can reduce the differences between the outputs of suboptimal decoders. In fact, it reduces uncertainty loss. Shrinking can also be achieved implicitly. The role of explicit loss term is that, in pseudo-supervised correction, the explicit loss term contributes more regularization during training, which helps stabilize the training process.

[0074] Understandably, to further enhance the utilization of unlabeled data, this invention introduces the high-performance SAM-Med3D as an additional supervisory branch to facilitate semi-supervised learning. Since SAM-Med3D contains rich and robust semantic information, this invention uses the prior knowledge of SAM-Med3D to enhance the semantic representation of unlabeled pheochromocytoma data. Specifically, this invention uses SAM-Med3D's predictions for unlabeled data as an additional supervisory signal to supervise the suboptimal decoder.

[0075]

[0076] This represents the predicted output of the suboptimal decoder. Representing the inference results of SAM-Med3D, by narrowing down To further optimize the suboptimal decoder and improve the performance of the segmentation model.

[0077] Furthermore, SAM-Med3D requires expert-provided cues, and this reliance on labor-intensive manual annotations hinders its clinical applicability. To address this issue, this invention utilizes the decoder output as location information to generate cues. Since coarse segmentation contains noise, filtering is necessary to retain reliable cues. Specifically, this invention selects voxels with higher foreground probability values ​​from pseudo-labels; higher foreground probabilities indicate more reliable predictions and play a greater role in pheochromocytoma prediction.

[0078]

[0079] Where T is the threshold for determining high-quality mask hints, and P i This represents the prediction of the pseudo-label on the i-th voxel. This is the mask hint information required by SAM-Med3D. This operation reduces false positive errors in mask prediction, retains high-probability voxels, and filters out low-confidence voxels that are prone to misclassification.

[0080] It is understood that the final loss function of this invention takes the following form:

[0081]

[0082] in, and γ and β are the supervised loss terms. Dice loss is a region-based loss function, and focal loss helps alleviate the problem of imbalanced samples. Both are widely used in segmentation tasks. and Two hyperparameters, ω and It is about balancing the loss of uncertainty. and The hyperparameters. It should be noted that... and It was applied to both labeled and unlabeled datasets. The pheochromocytoma segmentation model was trained based on the aforementioned final loss function, resulting in a well-trained pheochromocytoma segmentation model.

[0083] The following experiments verify the medical image segmentation method for pheochromocytoma proposed in this invention.

[0084] For the dataset, this invention evaluates a semi-supervised segmentation framework on the pheochromocytoma (PPGL) dataset. The PPGL dataset is a private 3D dataset collected by Hospital A, containing a total of 100 CT samples. The dataset was labeled by two experienced medical experts and then finalized by a senior expert. Furthermore, each sample had an isotropic resolution of 1×1×1 mm, with window width and window level set to 255 HU and 127.5 HU, respectively. To meet the input size requirements of SAM-Med3D, each sample was cropped to 128×128×128. We cropped the corresponding image based on the label location, preserving the lesion region while enlarging its edges, expanding it in three dimensions within the range of [10–20, 10–20, 5–10] voxels. Then, all samples were normalized to zero mean and unit variance. 80 samples were randomly selected as the training set, and the remaining 20 samples were used as the test set.

[0085] For evaluation metrics, this invention uses the DICE coefficient (DI), Jaccard (JA), average surface distance (ASD), and 95% Hausdorff distance (95HD). The DICE coefficient and Jaccard are used to measure pixel-level overlap between the segmentation map and the ground truth label; higher values ​​for these two metrics indicate better performance. The average surface distance (ASD) and 95% Hausdorff distance are used to evaluate boundary similarity; lower values ​​for these two metrics indicate better performance.

[0086] Based on this, this invention proposes a semi-supervised segmentation framework for pheochromocytoma segmentation. This framework mainly consists of a semi-supervised segmentation model and SAM-Med3D. The semi-supervised segmentation model processes both labeled and unlabeled data simultaneously, while SAM-Med3D provides additional supervision information and processes only unlabeled data. The semi-supervised segmentation model is an encoder-decoder model based on the popular V-Net architecture, consisting of a shared encoder and multiple independent decoders. To effectively address the challenges of pheochromocytoma's varying shape and size, non-fixed location, and low contrast, which lead to insufficient acquisition of key information, this invention designs a multi-scale attention module to enhance the extraction of key features, guiding the model to focus on the tumor region and reducing background noise interference. For pseudo-label correction, this invention sets up a dynamic uncertainty correction strategy, selecting the decoder output with the best performance from multiple decoder outputs as the pseudo-label to guide the suboptimal decoder. Dynamically competing for pseudo-labels among multiple prediction results can mitigate the inherent bias caused by a single model processing pseudo-labels of unlabeled pixels. Considering the potential noise interference in pseudo-labels that affects model training, this invention introduces an uncertainty correction method. This invention uses Kullback-Leibler (KL) divergence to estimate the confidence levels of labeled and unlabeled samples. Then, it enforces consistency between the predictions of the optimal decoder and the suboptimal decoder. Uncertainty estimation is a weighting factor used to control the importance of each sample in the consistency loss. This approach highlights more reliable predictions while assigning less weight to less reliable ones. Furthermore, addressing the deficiency of SAM-Med3D requiring manually provided cues, this invention generates high-quality mask cues to guide SAM-Med3D segmentation by filtering pseudo-labels obtained from the semi-supervised segmentation model. This method, utilizing positional information generated from predictions by multiple decoders as a source of cues, effectively solves the problem of labor-intensive manual cues.

[0087] For all experiments, the SGD optimizer was used to optimize the model, with a learning rate of 0.01 and a weight decay factor of 0.0001. The batch size was set to 4, containing two labeled patches and two unlabeled patches. Weights and φ were both set to 0.5, and both weights and φ were set as time-dependent Gaussian warm-up functions. The model was trained for 10k iterations. During training, random pruning, rotation, and flipping were used for data augmentation. In the testing phase, only the first decoder output was used as the final prediction. All experiments were run in the same environment: an NVIDIA 4090 hardware system, PyTorch 2.1.0 + CUDA 12.0 software, and a random seed of 1337.

[0088] The experimental results are as follows:

[0089] Table 1 shows the test results on the internal PPGL dataset, using 5% of the labels. The best results are shown in bold.

[0090]

[0091]

[0092] Table 2 shows the test results on the internal PPGL dataset, using 10% of the labels. The best results are shown in bold.

[0093]

[0094] Table 3 shows the test results on the internal PPGL dataset, using 20% ​​of the labels. The best results are shown in bold.

[0095]

[0096] Based on the data comparison in Tables 1, 2, and 3 above, the comprehensive experimental results on the internal pheochromocytoma dataset show that the method proposed in this invention is superior to other existing semi-supervised methods.

[0097] In summary, compared with the prior art, the pheochromocytoma medical image segmentation method of the present invention has at least the following beneficial effects:

[0098] 1. This invention proposes a semi-supervised segmentation framework that embeds SAM-Med3D prior knowledge, providing an automated solution for pheochromocytoma segmentation. The framework consists of a semi-supervised segmentation model and a SAM-Med3D model.

[0099] 2. To improve pseudo-label correction capabilities, this invention proposes a dynamic uncertainty correction strategy. This strategy selects pseudo-labels through competitive comparison and introduces uncertainty evaluation into the pseudo-labels, enabling the model to learn more reliably from unlabeled data. Simultaneously, this invention designs a multi-scale attention module in the semi-supervised segmentation model to guide the model to focus on specific domain features, thereby improving network performance.

[0100] 3. Comprehensive experimental results on an internal pheochromocytoma dataset demonstrate that the method proposed in this invention outperforms other existing semi-supervised methods.

[0101] In addition, such as Figure 5 As shown, one embodiment of the present invention also discloses a medical image segmentation device for pheochromocytoma, the device comprising:

[0102] Acquisition module 110 is used to acquire medical images of pheochromocytoma;

[0103] The segmentation module 120 is used to input the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation, and obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder. The multi-scale attention module is used to enhance the extraction of key features of pheochromocytoma. The pseudo-label is selected by competitive comparison among the outputs of multiple decoders, and the position information predicted by multiple decoders is used as the source of mask cue information for the SAM-Med3D segmentation model.

[0104] The pheochromocytoma medical image segmentation device of this invention is used to execute the pheochromocytoma medical image segmentation method in the above embodiments. Its specific processing procedure is the same as that of the pheochromocytoma medical image segmentation method in the above embodiments, and will not be described in detail here.

[0105] In addition, such as Figure 6 As shown, one embodiment of the present invention also discloses an electronic device, including: at least one processor 210; at least one memory 220 for storing at least one program; when the at least one program is executed by the at least one processor 210, it implements the pheochromocytoma medical image segmentation method as in any of the preceding embodiments.

[0106] In addition, one embodiment of the present invention discloses a computer-readable storage medium storing computer-executable instructions for performing the pheochromocytoma medical image segmentation method as described in any of the preceding embodiments.

[0107] The system architecture and application scenarios described in the embodiments of this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of system architecture and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0108] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0109] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0110] The terms “component,” “module,” “system,” etc., used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process or execution thread, and components may be located on a single computer or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems via signals).

Claims

1. A method for medical image segmentation of pheochromocytoma, comprising: Acquiring medical images of pheochromocytoma; The pheochromocytoma medical image is input into a pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder to enhance the extraction of key features of pheochromocytoma. False labels are selected through competitive comparison among the outputs of multiple decoders, and the positional information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model. The step of inputting the pheochromocytoma medical image into the pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image includes: segmenting the pheochromocytoma... A medical image is input to the encoder for feature extraction to obtain pheochromocytoma medical image features; the pheochromocytoma medical image features are input to multiple decoders to obtain multiple output results; a dynamic uncertainty correction strategy is used to select pseudo-labels from the multiple output results; the pseudo-labels are filtered to generate mask prompt information; the mask prompt information guides the SAM-Med3D segmentation model to perform image segmentation on the pheochromocytoma medical image to obtain the segmentation result of the pheochromocytoma medical image; wherein, the dynamic uncertainty correction strategy for selecting pseudo-labels from the multiple output results includes: selecting the optimal prediction result as a pseudo-label from the multiple output results through dynamic competition; introducing an uncertainty correction method to evaluate the selected pseudo-labels at the voxel level to obtain the confidence level of each voxel; and assigning a weight factor value to the corresponding voxel based on the confidence level of each voxel.

2. The method according to claim 1, characterized in that, The multi-scale attention module includes an MSA-H variant applied at high resolution and an MSA-L variant applied at low resolution. The process of inputting the pheochromocytoma medical image into the encoder for feature extraction to obtain pheochromocytoma medical image features includes: The pheochromocytoma medical image is input into the encoder to learn variable pheochromocytoma features; The variable pheochromocytoma features are input into the multi-scale attention module to extract multi-scale medical image features of the pheochromocytoma.

3. The method according to claim 1, characterized in that, The step of assigning a weight factor value to each voxel based on its confidence level includes: Increase the weight factor value of the voxels with high confidence; Reduce the weight factor value of the voxels with low confidence.

4. The method according to claim 1, characterized in that, The method further includes: The output of the optimal decoder is selected from the multiple decoder outputs and used as a pseudo-label to guide the suboptimal decoder.

5. The method according to claim 1, characterized in that, The training method for the pheochromocytoma segmentation model includes: Determine the final loss function, which is determined based on multiple supervised loss functions, multiple uncertain loss functions, and multiple balancing hyperparameters; The pheochromocytoma segmentation model is trained based on the final loss function to obtain the trained pheochromocytoma segmentation model.

6. A medical image segmentation device for pheochromocytoma, characterized in that, The device includes: The acquisition module is used to acquire medical images of pheochromocytoma; A segmentation module is used to input the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation, obtaining the segmentation result of the pheochromocytoma medical image. The pheochromocytoma segmentation model consists of a semi-supervised segmentation model and a SAM-Med3D segmentation model. The semi-supervised segmentation model includes a shared encoder and multiple independent decoders. A multi-scale attention module is set on the encoding path of the encoder to enhance the extraction of key features of pheochromocytoma. False labels are selected through competitive comparison among the outputs of multiple decoders, and the positional information predicted by multiple decoders serves as the source of masking cue information for the SAM-Med3D segmentation model. The step of inputting the pheochromocytoma medical image into the pre-trained pheochromocytoma segmentation model for image segmentation to obtain the segmentation result of the pheochromocytoma medical image includes: segmenting the pheochromocytoma medical image into a pre-trained pheochromocytoma segmentation model for image segmentation, obtaining the segmentation result of the pheochromocytoma medical image, including ... Pheochromocytoma medical images are input to the encoder for feature extraction to obtain pheochromocytoma medical image features; the pheochromocytoma medical image features are input to multiple decoders to obtain multiple output results; a dynamic uncertainty correction strategy is used to select pseudo-labels from the multiple output results; the pseudo-labels are filtered to generate mask prompt information; the mask prompt information guides the SAM-Med3D segmentation model to perform image segmentation on the pheochromocytoma medical image to obtain the segmentation result of the pheochromocytoma medical image; wherein, the dynamic uncertainty correction strategy for selecting pseudo-labels from the multiple output results includes: selecting the optimal prediction result as a pseudo-label from the multiple output results through dynamic competition; introducing an uncertainty correction method to evaluate the selected pseudo-labels at the voxel level to obtain the confidence level of each voxel; and assigning a weight factor value to the corresponding voxel based on the confidence level of each voxel.

7. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the pheochromocytoma medical image segmentation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing computer-executable instructions for performing the pheochromocytoma medical image segmentation method as described in any one of claims 1 to 5.

Images