124 results about "Computer-aided diagnosis" patented technology

The invention discloses a heterogeneous double-flow fusion method and system for diabetic retinopathy grading. The method comprises the following steps: obtaining an output result of diabetic retinopathy grading by utilizing a heterogeneous double-flow architecture; processing an input fundus image into images with different resolutions; extracting global context features from the low-resolution image by using a lightweight visual Transform model distilled by composite knowledge, and extracting local focus features from the high-resolution image by using a convolutional neural network model; performing interactive fusion on the global context features and the local focus features of the double-branch architecture through a symmetric bidirectional cross attention fusion module to obtain enhanced fusion feature representation; and finally, inputting the fusion features into a classifier, and outputting a severity grading result of the lesion. The method aims at improving the accuracy and robustness of hierarchical diagnosis through deep analysis of global information and local details, and can be applied to the medical fields of clinical computer-aided diagnosis, eye image analysis and the like.

The invention discloses a liver and gall lesion early screening system and method based on image fusion, and relates to the technical field of medical image processing and computer-aided diagnosis, and the method comprises the following steps: reconstructing a multi-modal image space-time coordinate system under a unified event time baseline, generating a respiratory displacement field and a magnetic sensitive pulse fingerprint, and constructing an artifact suspicion map; and performing anti-fact playback based on the artifact suspicion chart, performing frame-by-frame playback on the image acquisition sequence, quantifying artifact superposition tracks with consistent directions, and solidifying an artifact anchor point set. According to the method, space-time coordinates are constructed based on a unified event time baseline, a breathing displacement field and magnetic sensing pulse fingerprints are introduced, anti-fact playback, distortion kernel inference and residual decoupling are combined, artifact recognition and fusion intervention are achieved, and artifact closed-loop elimination is completed by judging threshold-driven fusion regulation and time reversal phase gating, so that the artifact recognition accuracy is improved. And the fused image authenticity is improved.

The invention discloses a brain tumor MRI image semantic segmentation method, and belongs to the technical field of medical image processing and computer-aided diagnosis. According to the invention, the problem of low segmentation precision obtained based on the existing brain tumor MRI image semantic segmentation technology is solved. The invention provides a semantic segmentation model which combines a U-Net framework with an LWD module, an MFDF module and various filters, the LWD module can keep information as much as possible in a down-sampling process, the MFDF module extracts operators by constructing new directional gradient features, and combines the operators in different directions by using dual-channel filtering, so that the semantic segmentation of the U-Net framework is realized. And the constructed multi-directional filter can respectively extract low-frequency and high-frequency characteristic direction information. The MFDF module transmits detail features from the coding module to the corresponding decoding module, so that spatial information including feature boundaries and textures is recovered, the precision of model segmentation is improved, and the method has good adaptability to the randomness of brain tumor shapes, sizes and boundaries. The method can be applied to brain tumor MRI image segmentation.

Systems and methods for performing one or more medical imaging analysis tasks on PCCT (photon-counting computed tomography) images are provided. Image acquisition parameters of a PCCT image acquisition device are determined for acquiring PCCT images. One or more PCCT images of an anatomical object of a patient acquired using the PCCT image acquisition device configured with the image acquisition parameters are received. One or more medical imaging analysis tasks analyzing the anatomical object are performed based on the one or more PCCT images using one or more machine learning based models. Results of the one or more medical imaging analysis tasks are output.

The invention discloses a thyroid ultrasound image diagnosis method based on deep learning, and the method comprises the following steps: collecting a thyroid ultrasound original image set, and carrying out the preprocessing; performing focus segmentation on the standardized thyroid ultrasound image set; performing morphological constraint and boundary refinement; calculating the blood flow direction, blood flow velocity and blood flow power of each thyroid focus area and neighborhood; generating a preliminary fusion feature map based on a feature adaptive deep learning network, and fusing the preliminary fusion feature map with the thyroid focus blood flow feature vector set; obtaining a thyroid focus detection list through thyroid focus benign and malignant discrimination branches; and generating thyroid focus structured diagnosis data records based on the thyroid focus detection list, and writing the thyroid focus structured diagnosis data records into a computer-aided diagnosis system. According to the method, deep learning and multi-modal blood flow features are fused, intelligent diagnosis of the thyroid focus is achieved, and the method has the advantages of being high in precision, high in interference resistance and structured in result.

The embodiment of the invention provides an image processing and computer-aided diagnosis method, electronic equipment, a storage medium and a program product, and the image processing method comprises the steps: obtaining a to-be-processed medical image which comprises the information of a plurality of target parts; performing first feature extraction on the medical image to obtain medical image features of the medical image; performing second feature extraction on the medical image features to obtain part identification information of each target part corresponding to the plurality of target parts in the medical image features; taking the part identification information of each target part as guidance, performing part feature extraction based on the medical image features, and obtaining part features of each target part; and predicting text features of the text based on the part features of the target parts and the detection results corresponding to the target parts to obtain prediction detection results of the multiple target parts. Therefore, the target part in the medical image can be accurately positioned and detected without depending on an external mask.

The invention discloses a variable convolution UNet tooth image segmentation method and system fusing an attention mechanism, and belongs to the technical field of medical image processing and computer-aided diagnosis. A tooth image data set is obtained and preprocessed; using the preprocessed data set to train a DSC-UNet segmentation model; the DSC-UNet segmentation model introduces a multi-modal feature fusion module and a deformable convolution module into a UNet framework; and segmenting an input tooth image by using the trained DSC-UNet segmentation model, and outputting a pixel-level segmentation result of the tooth structure. By constructing a multi-level feature interaction mechanism and a deformable convolution module, the capturing capability of the network on tooth edge details and the segmentation robustness of special-shaped teeth are remarkably improved, more reliable technical support is provided for an intelligent dental diagnosis and treatment system, and the method is particularly suitable for high-precision segmentation of complex tooth structures in oral panoramic X-ray films and CBCT images.

The invention relates to the field of computer-aided diagnosis, cross-modal fusion, image morphological operation and feature extraction related to computer vision and artificial intelligence, in particular to a method for predicting mild cognitive impairment in the progress stage of Alzheimer's disease, which comprises the following steps: S1, respectively inputting MRI (Magnetic Resonance Imaging), Clinical Table and PET (Positron Emission Tomography) data into respective encoders, and extracting initial image features and table features; s2, for the MRI branch and the PET branch, inputting respective image features and table features into a mixed channel attention module together, and calculating a mixed channel attention feature; s3, performing element-by-element multiplication on the mixed channel attention features obtained in the step S2 and original image features of corresponding branches to realize feature enhancement of channel dimensions; and S4, inputting the feature enhanced in the step S3 into a space attention module, and calculating a space attention feature. Structural MRI, functional PET and clinical table data are integrated for predicting the conversion from MCI to AD.

The invention discloses an intelligent auxiliary diagnosis and grading system for gastrointestinal tumor endoscopic images, which belongs to the technical field of medical image processing and computer-aided diagnosis and comprises a multi-modal image preprocessing and segmentation module, a lesion feature extraction and diagnosis module, a grading and depth evaluation module and a closed-loop feedback optimization module. The system receives white light, a narrow band and an amplified endoscopic image, adaptive segmentation is performed to obtain a lesion area, mucous membrane morphology, capillary and gland features are extracted, Paris classification, Vienna classification and infiltration depth evaluation are realized, and segmentation parameters are subjected to closed-loop optimization according to classification confidence. And intelligent auxiliary support is provided for early diagnosis and treatment decision of gastrointestinal tumors.

The invention relates to the technical field of computer-aided diagnosis, and discloses a block classification aided diagnosis system based on a self-prompt mechanism, which comprises the following steps: dividing a medical ultrasonic image into a plurality of image blocks, carrying out convolution and flattening processing to obtain a feature vector, extracting a preliminary global feature through a plurality of visual conversion blocks, and carrying out self-prompt on the image blocks; a fusion feature is obtained through a plurality of prompt attention conversion blocks, a self-prompt mechanism is introduced into the prompt attention conversion blocks, a learnable space prompt vector and a channel prompt matrix are introduced into the self-prompt mechanism, and a space mask is generated by calculating cosine similarity of the space prompt vector and a preliminary global feature. After spatial screening operation is carried out on the initial global features, channel guiding operation is applied through a channel prompt matrix; and performing weighted adjustment on the fused features to obtain final features, and obtaining benign and malignant classification results through a classification head. The feature modeling capability of the local fine-grained discriminative region in the image can be effectively improved, and the auxiliary diagnosis accuracy is improved.

The invention provides an intelligent assessment method for cervical lateral deviation based on multi-modal fusion and deep learning, and solves the technical problems of single data source, weak feature generalization ability and large assessment result error caused by only analysis of a static single frame in existing cervical lateral deviation assessment. The method comprises the steps of obtaining multi-modal data, performing preprocessing, attitude normalization and feature weighted fusion to obtain fused multi-scale features, performing mapping to generate a heat map, and performing peak retrieval on the heat map to output key points; tracking time sequence tracks of the key points, correcting abnormal key points, and collecting continuous multi-frame key points for key point detection to obtain target key points; and constructing a cervical vertebra trunk coupling mechanical model, dynamically optimizing a gravity line reference axis, correcting pelvic inclination attitude deviation, and optimally solving the optimal offset through a Lagrange multiplication method and a least square method. The method can be widely applied to the technical field of medical image analysis and computer-aided diagnosis.

The invention discloses a lightweight dynamic multi-scale segmentation method for a colon polyp. The lightweight dynamic multi-scale segmentation method comprises the following steps: S1, image input and data enhancement; s2, backbone network feature extraction: constructing a U-shaped encoder-decoder structure, wherein an encoder extracts multi-scale features through a dynamic multi-scale convolution module; s3, local and global feature enhancement: capturing local texture details through a regional detail aggregator, and obtaining global semantic information through a global context integrator; and S4, detection and segmentation: integrating the local details and the global semantic information by a decoder by adopting a hierarchical later fusion strategy, and outputting a final segmentation result. According to the method, the dynamic multi-scale convolution module is introduced, and the dynamic kernel selection and structure re-parameterization technology is used, so that the multi-scale feature extraction capability is greatly improved while the light weight of the model is kept; a regional detail aggregator and a global context integrator are introduced, so that local details and global semantics are effectively coordinated, and the segmentation precision of the polyp boundary is remarkably improved; compared with the prior art, the colon polyp segmentation method has the advantages that higher accuracy and intersection-union ratio are achieved on colon polyp segmentation tasks, and meanwhile, the colon polyp segmentation method has lower calculation complexity and parameter quantity and is suitable for real-time deployment of a clinical computer-aided diagnosis system.

The invention relates to the field of medical image processing and computer-aided diagnosis, in particular to an OCT (optical coherence tomography) retina layer and hydrops segmentation method based on a graph attention dual-decoder network, which comprises the following steps of: preparing and preprocessing data, acquiring OCT image data and generating corresponding segmentation labels and boundary labels; feature coding: processing the input image through a shared encoder network, and extracting and outputting feature maps of multiple levels; carrying out dual-path decoding and interaction, carrying out region segmentation and boundary segmentation by using two parallel decoders, and carrying out cross-path feature interaction through an attention mechanism during decoding; a graph attention mechanism is introduced into a region segmentation path, and information modeling is carried out on the deep features; and calculating a composite loss function according to the output of the decoder, and optimizing network parameters. According to the method, high-precision segmentation of the lesion OCT image is realized through a double-decoder multi-task architecture, cross-level space attention interaction and image attention global reasoning.

This invention relates to a grading method for diabetic retinopathy based on a local-global interactive dual-branch network, belonging to the fields of deep learning, medical image analysis, and computer-aided diagnosis. It includes: inputting the generated initial feature sequence into a global scanning module; using dilated reparameterized convolution to capture multi-scale structural prior information; inputting the feature map with structural prior information into a core dual-branch LoGo module; the local flow extracting fine-grained lesion features through depthwise separable convolution; the global flow generating spatially variable convolution kernels based on the global context through contextual hybrid dynamic convolution; and performing bidirectional cross-modulation through an adaptive multi-scale feature interaction aggregation module to sharpen local details for global semantics and filter local noise from the global context, ultimately outputting a disease severity grade. This invention explicitly simulates the cognitive strategies of clinicians, significantly improving grading accuracy while maintaining computational efficiency.

The invention provides a helicobacter pylori fluorescence image fine tuning detection method and device and a readable storage medium thereof, and belongs to the technical field of medical image processing and computer-aided diagnosis. The method aims at solving the problems that in the prior art, the false positive is high due to the fact that only a fluorescence signal is relied on, and a model is poor in generalization for different data environments. The method comprises the following steps: outputting a detection frame of a suspected target from a fluorescence image block through a detection module; determining a target analysis area containing a detection frame and a surrounding tissue area, and inputting the analysis area into a classification module to judge true and false positive; and when the classification result is confirmed to be wrong, only performing fine tuning training on a classifier in the classification module by utilizing target analysis area data corresponding to the error and combining a group of preset representative sample features. According to the method, the false positive rate is remarkably reduced by using surrounding organization information, and the generalization ability and scene adaptability of the model are improved through an efficient fine adjustment mechanism.

The invention discloses an Alzheimer's disease classification method based on sMRI and PET brain images. The Alzheimer's disease classification method is used for auxiliary diagnosis of the Alzheimer's disease. The method comprises the following steps that firstly, a feature extraction network is used for extracting image features of sMRI and PET brain images, and the feature extraction network enhances the spatial position information capturing capacity and the redundant information inhibiting capacity of the network by embedding a coordinate attention mechanism and space-channel reconstruction convolution; secondly, realizing complementary fusion of two image features by using a parallel interaction network based on proxy cross attention; and finally, classifying by using a full connection layer as a classifier. According to the method provided by the invention, the pathological information of sMRI and PET brain images is combined, the Alzheimer's disease classification accuracy is improved, a doctor can be assisted in diagnosis, and the method has a wide application prospect in the field of computer-aided diagnosis.

The invention discloses an ultrasonic shear wave elastography-based intrameniscus tearing quantitative evaluation method and system, and relates to the technical field of computer-aided diagnosis. The method is used for solving the problem of insufficient quantification of meniscus tearing detection. The method comprises the following steps: firstly, fixing a knee joint at a preset buckling angle, exciting shear waves along an annular fiber direction and a vertical direction, and collecting ultrasonic radio-frequency signals of an affected side meniscus and an uninjured side meniscus in different load states; then, a three-dimensional anisotropic elastic atlas is constructed based on the shear wave propagation velocity field, and spatial elastic distribution characteristics of the tissue are obtained; thirdly, calculating an elastic continuity index, a spatial variation coefficient and an abnormal region morphological parameter, and quantitatively representing the continuity and local tearing characteristics of the tissue structure; and finally, inputting the multi-dimensional features into a multi-branch neural network, fusing the features through a cross-branch attention mechanism, and outputting a meniscus tearing quantitative evaluation value, thereby realizing intelligent and objective evaluation of the tearing degree, and providing a basis for clinical diagnosis and postoperative rehabilitation.

This invention discloses a multi-label classification method for medical images based on spatiotemporal integration and adaptive normalization, belonging to the field of medical image processing and computer-aided diagnosis technology. The invention proposes an improved scheme. Firstly, it constructs a dynamically balanced data input stream through frequency inverse sampling and cosine annealing strategies. Secondly, it introduces group normalization to replace batch normalization in the feature extraction network, solving the training convergence problem under limited GPU memory. During training, it employs a focal loss function to mine hard-to-classify samples. In the inference stage, the invention constructs a spatiotemporal integration mechanism based on multi-view space transformation and multi-stage model weight fusion, combined with an adaptive threshold decision algorithm based on maximizing F1 scores to output diagnostic results. This invention effectively improves the recognition accuracy of key pathological features such as cardiac hypertrophy and edema, while significantly improving the recall rate of small lesions, exhibiting high robustness and clinical auxiliary value.

The application relates to a light wavelet enhancement fusion medical image segmentation method, and belongs to the technical field of medical image processing and computer-aided diagnosis. The core of the method is to construct a light WEF-Net network, which comprises an encoder, a bottleneck layer and a decoder. The encoder adopts a dual-domain perception module, extracts complementary features from the frequency domain and the time domain through wavelet transformation and convolution, and simultaneously extracts complementary features from the frequency domain and the time domain. The bottleneck layer is designed with a feature aggregation Kolmogorov-Arnold transformation module, which is used for efficiently fusing multi-scale semantic information and enhancing the nonlinear modeling capability. The network level also integrates a sawtooth rolling feature fusion module, which enhances the global continuity of the features through the channel rolling and spatial scanning mechanism to improve the integrity of the boundary segmentation. Experiments show that the method realizes excellent performance on multiple public medical image datasets, while maintaining low parameter quantity and low computational quantity, and significantly improves the segmentation precision.

The invention discloses a computer-aided diagnosis method and system for Alzheimer's disease analysis, and relates to the technical field of medical information processing, and the key points of the technical scheme are as follows: the method comprises the following steps: obtaining a T1 weighted brain magnetic resonance image of a subject; dividing the two-side sea horse tooth-shaped gyrus areas; image omics features are extracted from the subregions; selecting a group of discriminative feature combinations containing texture features from the right dentate loop and first-order statistical features from the left dentate loop through a multi-layer screening strategy; and inputting the feature combination into a pre-trained machine learning model to generate a diagnosis classification result. According to the method, high-precision and high-sensitivity diagnosis of the Alzheimer's disease is realized by using non-invasive standard MRI data and capturing the microstructure change of dentate gyrus, and the diagnosis result has strong correlation with key plasma biomarkers and cognitive level, so that the method has important clinical application value.

The invention provides a breast lump segmentation method based on channel-guided double-pooling multi-scale space attention, and aims to solve the problems of small area, low contrast, fuzzy boundary and the like of breast lumps in an X-ray image, the network extracts full-view image features and compresses space dimensions through a deep residual encoder; and realizing fusion of low-level details and high-level semantic features in the decoder by means of jump connection. A double-pooling gating mechanism is built in the decoder, channel guide weights are generated in parallel through global average pooling and maximum pooling, lesion significant features are screened in a self-adaptive mode, and redundant backgrounds are restrained; the multi-scale space attention module captures multi-scale space information through multi-branch large-kernel separable convolution, generates a space attention graph, combines the space attention graph with channel weights element by element, and accurately focuses a lesion area and a boundary. Experiments show that the Dice coefficients of the method on INbreast, CBIS-DDSM and private In-home data sets respectively reach 90.94%, 80.60% and 84.50%, the method is superior to a mainstream method, the segmentation precision and generalization ability are improved, and reliable support is provided for early screening and computer-aided diagnosis of breast cancer.

The application relates to a chronic kidney disease prediction method based on clinical information graph representation, belongs to the technical field of computer-aided diagnosis of chronic kidney disease (CKD), and aims to solve the problem of missed diagnosis caused by the fact that the kidney function index of early CKD is not obvious. The incidence of CKD is high, the early symptoms are hidden, and the disease is easy to be missed, thus developing into end-stage renal disease. Therefore, the application provides a chronic kidney disease prediction method based on clinical information graph representation. The method first extracts fundus image and clinical index features; the clinical index is fused into a clinical index joint feature through text embedding and numerical feature fusion, and the joint feature is fused with the fundus image feature through cross-modal attention; the similarity between subjects is calculated based on the fused feature, and a subject relationship graph is constructed by using an adaptive dynamic threshold mechanism; finally, a hybrid graph neural network is used for graph representation learning, pathological similarity between subjects is mined, and accurate prediction of chronic kidney disease is realized. The application can improve the detection rate of early CKD and is applied to non-invasive early screening and risk early warning.

The invention discloses a breast cancer MRI image enhancement and focus automatic identification method based on an attention mechanism, and relates to the technical field of data processing and computer-aided diagnosis. The method comprises the following steps: acquiring a multi-time-point mammary gland MRI image sample and performing preprocessing; performing random window enhancement and random channel exchangeable enhancement on the preprocessed sample, inputting the enhanced image sample into an improved convolutional neural network, and performing deep metric learning loss function optimization based on proxy NCA to extract focus embedding features; according to the focus embedding features, constructing a moment matching game framework with focus state distribution marked by experts as a target; and according to a moment matching game framework, adopting an inverse reinforcement learning strategy without reinforcement learning, carrying out iterative optimization on parameters of the lesion recognition model through a regret-moment-free matching algorithm, carrying out lesion property judgment according to the optimized model, and outputting a judgment result and a corresponding confidence score. Therefore, MRI image adaptive enhancement and accurate focus automatic identification are realized.

A system for intelligent surveillance of attention and recognition of computer-aided diagnosis system outputs. The system can include an endoscope, a computer-aided diagnostic module, a camera, a memory, and a controller. The endoscope can include an elongated member that can include a distal portion and a process camera attached to the distal portion. The process camera can capture a video stream during a procedure. The computer-aided diagnostic module can be configured to detect an abnormality within the video stream using a diagnostic algorithm and transmit a signal. The controller can be configured to determine a gaze location of the doctor during the procedure using a gaze algorithm, determine whether the doctor looked at the detected abnormality by comparing the signal from the computer-aided diagnostic module and the gaze location of the doctor, and trigger a countermeasure based on determining that the doctor did not look at the detected abnormality.

The embodiment of the invention provides a dynamic function network adaptive construction method for physiological signal analysis and related equipment, and belongs to the technical field of biomedical signal processing and computer-aided diagnosis. The method comprises the following steps: inputting a multi-channel physiological signal and extracting a preset frequency band; performing time sequence segmentation on the signal by adopting a sliding window; calculating a phase locking value matrix among all channels for each time window; carrying out personalized sparsification processing on each connection matrix by adopting a self-adaptive threshold value method; and constructing a dynamic graph sequence based on a sparse result, wherein node features adopt weighting centrality. According to the method, the time-varying information of the brain function network is completely reserved through the dynamic graph sequence, the noise robustness and personalized processing are realized through the adaptive threshold, the learning efficiency of the subsequent graph neural network is enhanced through the structure-related node features, the accuracy of disturbance of consciousness classification is remarkably improved, and the classification efficiency of the disturbance of consciousness is improved. And the method has the universality of expanding to other physiological signal analysis and industrial monitoring fields.

A computer-implemented diagnostic-assistance system for medical applications, comprises: an artificial intelligence neural network configured to classify images of an obtained image dataset according to a set of classes; a confidence module configured to generate a confidence measure associated with each of the classified images; a tagging module configured to generate, for the patient, a diagnostic signal based on the generated confidence measures associated with the classified images, wherein the diagnostic signal for the patient is tagged as conclusive if a processed combination of the confidence measures fulfills a condition and tagged as inconclusive if the processed combination of the confidence measures does not fulfill the condition; and an output interface configured to output the diagnostic signal, wherein if the diagnostic signal is conclusive the classification result is released, and if the diagnostic signal is inconclusive an additional diagnostic analysis is triggered.

This invention discloses a heterogeneous two-stream fusion method and system for grading diabetic retinopathy (DR), comprising obtaining the DR grading output using a heterogeneous two-stream architecture: processing the input fundus image into images of different resolutions; extracting global contextual features from the low-resolution image using a lightweight visual Transformer model distilled from composite knowledge, and extracting local lesion features from the high-resolution image using a convolutional neural network model; interactively fusing the global contextual features and local lesion features of the two-branch architecture through a symmetrical bidirectional cross-attention fusion module to obtain an enhanced fused feature representation; and finally inputting the fused features into a classifier to output the DR severity grading result. This invention aims to improve the accuracy and robustness of grading diagnosis through in-depth analysis of global information and local details, and can be applied to medical fields such as clinical computer-aided diagnosis and ocular image analysis.

The invention provides a medical image-based computer-aided diagnosis method, system and device, a computer storage medium and a computer program product. The method comprises the following steps: acquiring original DICOM data obtained based on a tomography technology; constructing three-dimensional volume data based on the original DICOM data; performing multi-plane reconstruction based on the three-dimensional volume data to generate an MPR image queue; inputting the MPR image queue into a deep learning model to obtain a multi-scale feature map corresponding to each MPR image in the MPR image queue; using a multi-plane interaction module to carry out cross-plane association fusion on the multi-scale feature map to obtain a multi-plane fused feature map; and performing global pooling and full-connection coding on the multi-plane fused feature map by using a feature fusion and coding module, and outputting a one-dimensional diagnosis vector for auxiliary diagnosis. By adopting the scheme, the original medical image data can be efficiently converted into the diagnosis vector, and automatic computer-aided diagnosis is realized.

The application discloses a skeleton quality driven tubular structure segmentation closed loop optimization method and system, which is applied to the technical field of biomedical image processing and computer aided diagnosis, and the method comprises the following steps: using a trained segmentation model to infer medical volume data to obtain a segmentation mask; skeletonizing the segmentation mask to extract a skeleton graph; performing topological defect detection on the skeleton graph to obtain a structured defect report; reversely mapping three-dimensional coordinates in the report back to a voxel space, taking each defect point as a center, a preset radius as a range, and generating a defect density weight graph according to a severity score; and using the weight graph as a spatial weighting parameter of a loss function to optimize the segmentation model; and the application reversely maps the topological defects detected by skeletonization into a weight graph and integrates the weight graph into a loss function for closed loop iteration, realizes directional repair on high-occurrence areas such as fractures and false branches, and improves the conduction efficiency of segmentation improvement to skeleton quality.

The application belongs to the field of intelligent medical computer-aided diagnosis application, and relates to a colon polyp segmentation method based on deep learning fusion and collaborative learning enhancement. The method comprises a feature extraction model, a fusion module and multi-view collaborative learning. The feature extraction model is divided into two branches. One branch uses DeiT-Small to extract global feature information and establish the correlation between each pixel. The other branch uses HardNet-MSEG to extract local feature information and obtain more low-level detail information. In order to improve the segmentation accuracy of small target colon polyp images, based on the public colon polyp image dataset and the initial deep learning single-branch segmentation method, a deep learning technology fusion method is proposed, and multi-view collaborative learning is used to enhance colon polyp segmentation. Compared with the feature extracted by a single deep learning method before improvement, the feature is richer, and the information omission defect of a single branch is compensated.