Sun Yat-sen Memorial Hospital to Transfer Non-invasive Axillary Lymph Node Status Detection Technology for RMB 100,000

Recently, Sun Yat-sen Memorial Hospital of Sun Yat-sen University released a public notice on the transformation of scientific and technological achievements. The university intends to transfer its intellectual property rights through listed transactions,“Method, Apparatus, Device, and Storage Medium for Detecting Status Characteristics of Axillary Lymph Nodes”Relevant patents are transferred to industry partners.

The patent ownership is jointly held by Sun Yat-sen Memorial Hospital, Sun Yat-sen University, and the assignee. Sun Yat-sen Memorial Hospital, Sun Yat-sen University, intends to transfer its equity interest in the ownership of this invention patent. The transfer amount is100,000 yuan. The inventor of this patent isProfessor Xiang Zhang and his team。

The present invention relates to a method, apparatus, device, and storage medium for detecting status characteristics of axillary lymph nodes, the core of which lies in leveragingNon-invasive Technology, to identify the metastasis-related status characteristics of axillary lymph nodes in breast cancer patients preoperatively, thereby addressing the complications caused by existing invasive examinations while improving detection accuracy.

This technology can be applied to the preoperative assessment of axillary lymph node status in breast cancer patients, helping physicians determine whether invasive procedures such as sentinel lymph node biopsy (SLNB) or axillary lymph node dissection (ALND) are necessary, thereby reducing unnecessary trauma. Meanwhile, it provides a precise basis for formulating treatment plans and predicting prognosis, demonstrating significant application prospects in the clinical diagnosis of breast tumors.

Axillary lymph node status serves as a critical basis for formulating treatment plans and determining prognosis in breast cancer patients. With the widespread adoption of breast cancer screening and the continuous upgrading of diagnostic and therapeutic needs, there is a sustained surge in clinical demand for non-invasive and precise assessment of axillary lymph node status. However, current detection technologies suffer from numerous significant drawbacks, which severely impair clinical diagnostic and therapeutic efficiency as well as patient experience, failing to meet the urgent market demand for safe and efficient testing solutions.

Traditional detection technologies face a core medical bottleneck: on one hand, these techniques rely on invasive procedures, imposing a significant burden on patients. Currently, the mainstream clinicalSentinel Lymph Node Biopsy(SLNB) andAxillary Lymph Node Dissection(ALND) are invasive procedures that are highly likely to cause postoperative complications such as upper limb edema, sensory neuropathy, and limited mobility, which severely affect patients' quality of life. Furthermore, some low-risk patients are subjected to unnecessary surgical trauma.

On the other hand, there are deficiencies in the accuracy and comprehensiveness of testing. Invasive examinations may be influenced by factors such as sampling range and the subjectivity of pathological interpretation, potentially leading to missed diagnoses or overdiagnosis.

Furthermore, this examination cannot comprehensively assess key indicators such as the probability and number of metastases preoperatively, making it difficult to provide precise evidence for clinical decision-making.

From the perspective of clinical application scenarios, existing technologies still exhibit significant limitations in practical utility. Certain non-invasive auxiliary diagnostic methods (such as ultrasound and conventional MRI) can only provide morphological information, making it difficult to capture changes in the lymph node microenvironment and metastatic characteristics, resulting in relatively low sensitivity and specificity.

A minority of imaging-based detection schemes fail to integrate multidimensional imaging information, resulting in insufficiently comprehensive correlative analysis between the primary tumor and the axillary region, which limits the accuracy of assessing metastatic burden and the metastatic status of non-sentinel lymph nodes.

Meanwhile, traditional testing procedures lack standardized protocols. Clinical variations exist in invasive examinations regarding surgical timing and sampling methods, while the interpretation of images from auxiliary tests relies on physicians’ experience, further increasing the uncertainty of test results.

Moreover, existing technologies fail to strike a balance between diagnostic safety and clinical efficiency. Invasive examinations require a post-procedure recovery period, which prolongs patients’ waiting time for treatment; meanwhile, inefficient non-invasive testing protocols may lead to repeat examinations, thereby increasing healthcare costs and the burden on patients.

These issues have led to difficulties in detecting axillary lymph node status.“Invasive procedures carry risks, non-invasive methods lack precision, and clinical decisions lack evidence.”...dilemma. There is an urgent clinical need for a non-invasive, multi-dimensionally integrated, and precise comprehensive diagnostic solution to address the pain points throughout the entire process from preoperative assessment to treatment decision-making.

To address the industry pain points in axillary lymph node detection for breast cancer, namely “high invasiveness risk, insufficient accuracy, and reliance on a single decision-making basis,” the joint R&D team has successfully launched the technology titled “Method, Device, Equipment, and Storage Medium for Detecting Status Characteristics of Axillary Lymph Nodes.”

This technology, based on“Non-invasive Testing + Deep Learning Multi-task Fusion”as its core advantage, it has built a system coveringImage Acquisition, Region Extraction, Feature Encoding, and State DetectionEnd-to-End Solutions, thoroughly overcoming the limitations of traditional invasive examinations and providing a safer, more precise, and efficient new approach to preoperative assessment in the field of breast diagnosis and treatment.

This technology achieves the detection of axillary lymph node status"Non-invasive Transformation", fundamentally avoiding the medical risks associated with traditional invasive examinations. Traditional methods such as sentinel lymph node biopsy (SLNB) and axillary lymph node dissection (ALND) inevitably lead to postoperative complications, including upper limb edema, sensory nerve disorders, and limited mobility. In contrast, this technology requires only breast magnetic resonance imaging (MRI) scans from the subject to complete preoperative assessment through non-invasive means. The detection process eliminates the need for surgical or needle biopsy procedures, thereby sparing patients from unnecessary trauma and pain, obviating the postoperative recovery period, shortening treatment wait times, and significantly enhancing the patient’s healthcare experience.

Meanwhile, the non-invasive modality is suitable for patients with various types of breast cancer, particularly offering a safe and feasible testing option for those who cannot tolerate invasive examinations, such as the elderly and individuals with frail constitutions, thereby significantly expanding the scope of preoperative assessment.

Leveraging end-to-end innovations in image processing, feature extraction, and model optimization, the technology has established a detection framework based on “multimodal information fusion + precise regional focusing + deep feature mining,” significantly improving assessment accuracy.

First, conduct multimodal image stitching and normalization to lay a solid foundation for precise detection.This technique acquires breast magnetic resonance imaging (MRI) scans from at least two different imaging modalities, such as T1-weighted, T2-weighted, diffusion-weighted imaging (DWI), dynamic contrast-enhanced (DCE), and apparent diffusion coefficient (ADC) sequences. These images are concatenated along the channel dimension to generate target detection images that incorporate multidimensional information, including morphological and molecular-level features.

Prior to splicing, the following steps are performed:Spatial Normalization Alignment(ensuring that the same spatial point across different images corresponds to the same pixel) andIntensity Normalization(unified pixel intensity range). This measure effectively mitigates detection errors caused by image information bias, enabling the target images to more accurately reflect the true status of the breast and axillary lymph nodes.

Second, perform precise regional extraction to reduce interference from irrelevant information.By employing a sliding-window approach to crop the image, the target detection image is divided into multiple sub-images. A pre-defined region-of-interest (ROI) detection model is then applied to accurately identify and extract the primary tumor region (containing multiple mass images) and the axillary regions (covering both left and right axillary images). Subsequent analysis focuses on these core regions, thereby excluding interference from irrelevant image information, enhancing the specificity of feature extraction, and improving detection accuracy at the source.

Third, deep learning encoding and multi-task fusion techniques are employed to comprehensively mine feature information.Innovatively adopting a multi-encoder collaborative working mode, deep feature encodings of the primary tumor and axillary regions are extracted separately by leveraging shared-weight encoders combined with attention models. Meanwhile, the distance discrepancy between the left and right axillary region encodings is utilized to enhance feature discriminability.

The multi-encodings extracted by the first, second, third, and fourth encoders are respectively input into three multilayer perceptrons to jointly detect the probability of sentinel lymph node metastasis, the number of metastases (1–2 indicating low burden, ≥3 indicating high burden), and the probability of non-sentinel lymph node metastasis, thereby achieving“Metastasis status, metastatic burden, and extent of spread”Simultaneous Assessment of These Three Core Indicators.

End-to-end multi-task training can further optimize model parameters, enabling more comprehensive feature extraction and more reliable detection results, thereby providing multidimensional, precise evidence for clinical decision-making.

This technology demonstrates significant advantages in clinical utility, standardization, and scalability, making it fully adaptable to large-scale hospital diagnosis and treatment scenarios.

AtTesting Efficiency and WorkflowIn this regard, by leveraging designs such as sub-image partitioning and shared-weight encoders, the technology optimizes computational efficiency while ensuring detection accuracy. It eliminates the need for complex and time-consuming operational procedures, enables seamless integration with existing breast MRI equipment in hospitals, and does not incur additional costs in terms of diagnostic workflows or time.

InStandardization and StabilityIn this regard, the technology specifies core parameters for each stage—including image acquisition, preprocessing, encoding models, and threshold determination (e.g., setting the sliding window size to 64×64×64 and the default probability threshold to 50%)—thereby mitigating result variability associated with traditional detection methods that rely on physician experience. Validated across multiple image datasets, the detection results demonstrate high consistency and stability, effectively ensuring uniform detection standards across different medical institutions and operators.

InPromotion of ApplicabilityIn this regard, the technology provides a comprehensive solution encompassing detection methods, dedicated devices, hardware equipment, and storage media. It can be directly integrated into existing hospital diagnosis and treatment systems without requiring large-scale equipment modifications. The device adopts a modular design—comprising three major modules: image stitching, region detection, and state feature detection—facilitating maintenance and upgrades. The hardware can utilize common intelligent terminals such as standard computers and servers, thereby lowering the threshold for hospitals to adopt this technology and promoting its rapid application across medical institutions at all levels.

Furthermore, this technology demonstrates significant advantages in healthcare cost containment and the translation of clinical value. By employing a non-invasive approach, it reduces medical expenditures associated with invasive surgeries and postoperative care, while simultaneously avoiding the costs of ineffective treatments resulting from missed or overdiagnoses, thereby alleviating the financial burden on both patients and health insurance systems.

The comprehensive status features provided by this technology—including metastasis probability, number of metastases, and risk of non-sentinel lymph node metastasis—enable physicians to precisely formulate treatment plans. For instance, patients with low-burden metastasis can avoid axillary lymph node dissection (ALND), a more invasive procedure, while those with high-burden metastasis can have their treatment pathways directly optimized. This facilitates the implementation of “precision medicine” and significantly enhances the overall quality and efficiency of breast cancer diagnosis and treatment.

Currently, in the field of non-invasive detection of axillary lymph node status, a competitive landscape has emerged characterized by “medical universities and colleges conducting technological breakthroughs + AI healthcare companies driving commercial implementation.” Similar technologies are primarily focused onInfrared Thermography, AI Analysis of Medical ImagingThese two major approaches. Core enterprises and research teams are actively advancing technical validation and clinical translation; however, significant differences exist in terms of detection dimensions, applicable scenarios, and commercialization progress.

Google Health(Google Health)’s Lyna lymph node metastasis-assisted diagnostic model was developed based on convolutional neural networks (CNN) and integrated into the Augmented Reality Microscope (ARM) system. This system uses a camera to capture images of lymph node biopsy tissues, and the AI model automatically identifies metastatic cancer foci within the tissue, overlaying annotations in an augmented reality format to assist pathologists in making diagnoses.

The core of this technology lies in the precise identification of pathological slides, which can reduce the risk of missed diagnoses in manual diagnosis and is particularly suitable for the auxiliary analysis of biopsy samples.

GE Healthcare(GE Healthcare)’s Edison platform, paired with an AI algorithm for lymph node detection, uses the Edison platform as the core for AI integration, embedding the lymph node detection algorithm into its full range of imaging equipment (CT, MRI, ultrasound, etc.). By leveraging multi-modal image feature extraction and automated organ segmentation technology, it assists physicians in identifying signs of axillary lymph node metastasis. The algorithm is deeply optimized for GE’s Signa series MRI and Revolution series CT scanners.

Following the acquisition of Mim Software in 2024, capabilities in multimodal image integration and oncologic imaging analysis were further enhanced. This algorithm reduces radiologists' image interpretation time by 30%–40% and demonstrates stable specificity in differentiating benign from malignant lymph nodes on MRI equipment; however, it exhibits strong device dependency. The AUC value on GE’s proprietary equipment is approximately 0.62, which is lower than its performance when adapted to third-party devices.