Intelligent navigation system and method for breast-conserving surgery perforator flap based on CT three-dimensional blood vessel reconstruction and AI dynamic blood flow analysis

Abstract

The application discloses a breast-conserving surgery perforator flap intelligent navigation system and method based on CT three-dimensional blood vessel reconstruction and AI dynamic blood flow analysis, and comprises an image processing module. The image processing module generates a perforator blood vessel three-dimensional model with an accuracy of plus or minus 0.1 mm based on high-resolution CT thin-layer scanning data with a layer thickness of less than or equal to 0.625 mm through Mimics software. The application significantly improves the postoperative outcome of patients through multidimensional technology cooperation. In terms of complication control, the incidence of complications in the donor area is reduced from 22% in the traditional method to 3% (a reduction of 86%), and the expansion of trauma caused by perforator variation is avoided. In terms of aesthetic effect, for the breast-conserving surgery of medial quadrant breast cancer, the lateral thoracic artery perforator flap (LTAP) is combined with AI volume compensation. After the surgery, the postoperative breast upper pole depression rate is reduced from 35% to 5%, the BREAST-Q score of the patient is increased from 68 to 89, the requirements of 'tumor radical treatment' and 'functional aesthetics' are fully considered, the postoperative life quality of the patient is greatly improved, and the humanistic concept of modern tumor treatment is met.

Description

Technical Field

[0001] This invention relates to the field of breast-conserving surgery technology, and in particular to an intelligent navigation system and method for perforator flaps in breast-conserving surgery based on CT three-dimensional vascular reconstruction and AI dynamic blood flow analysis. Background Technology

[0002] Breast cancer is one of the most common malignant tumors among women worldwide. As cancer treatment approaches the balance between radical resection and functional aesthetics, breast-conserving surgery has become the preferred procedure for early-stage breast cancer. Its core requirement is to achieve complete tumor resection (negative margins) while simultaneously reconstructing breast morphology through perforator flap transplantation, thus reducing postoperative deformity rates. Perforator flap technology relies on precise localization and blood flow assessment of subcutaneous micro-perforator vessels (often <1mm in diameter). However, its clinical application still faces multi-dimensional technical bottlenecks. Existing methods are insufficient to meet the surgical requirements of "precision, dynamism, and integration." Specific issues include:

[0003] 1. Insufficient precision in preoperative image reconstruction leads to a high rate of missed detection of minute perforators.

[0004] Traditional breast-conserving surgery relies on conventional CT angiography (CTA) or ultrasound for perforator vessel assessment, which has significant limitations: On the one hand, the slice thickness of conventional CT scans is mostly 1-2mm, making it difficult to clearly visualize tiny perforator vessels with a diameter of <0.5mm, resulting in a false negative rate of over 30%. This leads to changes in the flap plan during surgery due to perforator vessel localization errors (such as changing from a perforator flap to a more invasive latissimus dorsi myocutaneous flap); on the other hand, current technology lacks multimodal image fusion capabilities, relying solely on single CT or ultrasound data, and cannot correct vessel localization errors through complementary information. As a result, the accuracy of the perforator vessel tree model can only reach ±1.2mm, which is insufficient to meet the requirements of millimeter-level surgical operations.

[0005] 2. Static blood flow analysis and surgical decision-making lead to insufficient postoperative risk prediction.

[0006] Current perforator flap assessments mostly use static indicators (such as vessel diameter and direction of travel), which cannot simulate the hemodynamic changes after flap transplantation. Firstly, the lack of time-series analysis makes it impossible to predict the dynamic evolution of flap blood perfusion within 72 hours post-surgery (such as pressure decrease and flow rate slowdown), resulting in low accuracy in assessing flap necrosis risk (AUC is only 0.75). Secondly, surgical decision-making lacks a multi-objective collaborative mechanism. Traditional approaches only focus on tumor margins or single aesthetic indicators, failing to integrate the three requirements of "tumor safety (margin ≥2mm)," "minimum donor site damage (scar <8cm)," and "aesthetic symmetry (Δ volume ≤10%)." This easily leads to contradictions such as "safe margins but deformed morphology" or "aesthetically satisfactory but high donor site complication rate." Postoperative local recurrence rate remains around 4.5%, and the incidence of donor site complications reaches 22%.

[0007] 3. Intraoperative navigation lacks dynamism and safety, resulting in a high risk of perforator injury.

[0008] Breast-conserving surgery requires real-time adjustment of the flap harvesting area to adapt to anatomical variations. However, existing navigation technologies have significant shortcomings: First, augmented reality (AR) navigation devices (such as early HoloLens models) lack dynamic error compensation mechanisms. Affected by intraoperative tissue traction and respiratory movements, navigation errors can reach 3-5 mm, making it impossible to accurately guide perforator anastomosis. Second, there is a lack of active safety protection measures. There is no tactile feedback when surgical instruments approach perforator vessels, resulting in a high rate of accidental injury to perforator vessels and a perforator anastomosis success rate of only 85%. Third, the patency of perforators cannot be verified in real time during surgery, relying solely on the surgeon's experience. If perforator occlusion is found postoperatively, a second surgery is required for repair, increasing patient trauma.

[0009] 4. The entire process is inefficient and difficult to standardize, limiting its adoption in primary hospitals.

[0010] Traditional breast-conserving surgery relies on manual preoperative planning by experts, requiring the integration of steps such as image interpretation, flap design, and risk assessment, taking up to 120 minutes. Furthermore, the quality of the plan is significantly affected by the surgeon's experience (the structural similarity index (SSIM) of different doctors' plans is <0.85). In addition, intraoperative decisions depend on real-time discussions, often with delays exceeding 10 seconds, which can easily interrupt the surgical process. Moreover, the lack of a unified "preoperative-intraoperative-postoperative" data management system makes it impossible to establish standardized operating procedures. As a result, due to the high technical threshold and long learning curve, primary hospitals find it difficult to routinely perform complex perforator flap breast-conserving surgery, thus limiting the widespread application of the technology. Summary of the Invention

[0011] To overcome the shortcomings of the prior art, one of the objectives of this invention is to provide an intelligent navigation system and method for perforator flaps in breast-conserving surgery based on CT three-dimensional vascular reconstruction and AI dynamic blood flow analysis.

[0012] One of the objectives of this invention is achieved through the following technical solution:

[0013] A breast-conserving surgery perforator flap intelligent navigation system includes: an image processing module, which generates a three-dimensional model of perforator vessels with an accuracy of ±0.1mm based on high-resolution CT thin-slice scan data with a slice thickness ≤0.625mm using Mimics software, and integrates a multimodal image fusion unit. The multimodal image fusion unit adopts an adaptive weighted dynamic fusion algorithm, which fuses CT image data and ultrasound contrast data to correct the localization of small perforator vessels with a diameter <0.5mm through dual constraints of gray value matching and vascular morphology features, and outputs a corrected perforator vessel tree model. The structural similarity index (SSIM) of the preprocessed image is ≥0.95.

[0014] The AI ​​decision-making module incorporates a hemodynamic prediction and optimization system based on an LSTM-DRL hybrid architecture. The LSTM neural network layer simulates the temporal changes in blood perfusion after perforator flap transplantation, including dynamic curves of blood pressure, flow velocity, and oxygen saturation, and predicts the flap survival rate (AUC≥0.92) 72 hours post-surgery. The deep reinforcement learning (DRL) optimization unit uses tumor invasion depth and HER-2 expression status as clinical feature parameters to construct a three-dimensional reward function that includes tumor safety boundaries (≥2mm), BREAST-Q aesthetic scores, and donor site damage. It solves for the Pareto optimal solution using an improved NSGA-III algorithm, outputting a personalized breast-conserving reconstruction plan with a volume compensation rate ≥95%.

[0015] The dynamic navigation module includes an augmented reality device, a bivariate Kalman filter unit, and a graded force feedback device. The augmented reality device has a refresh rate of ≥60Hz and a spatial positioning error of ≤2mm, and is used to project the perforator vessel path and flap cutting range to the surgical field in real time. The bivariate Kalman filter unit performs dynamic error correction through an image drift-tissue deformation coupling model. The graded force feedback device is equipped with a nonlinear resistance algorithm based on the vessel diameter. When the distance between the surgical instrument and the perforator vessel with a diameter <1mm is ≤2mm, gradient resistance feedback is triggered. The resistance increases exponentially as the distance decreases, and the dynamic navigation module ensures that the perforator vessel anastomosis success rate is ≥98%.

[0016] A method for planning a perforator flap in breast-conserving surgery includes the following steps:

[0017] S1. Preoperative image acquisition: A high-resolution CT scanner with a slice thickness of ≤0.625mm was used to scan the patient's breast and flap donor area. The scanning range covered the bilateral breasts, chest wall, abdominal wall and back area to ensure that potential flap donor area vessels such as the thoracodorsal artery, lateral thoracic artery and inferior epigastric artery were included. At the same time, contrast agent peak triggering technology was used to capture the imaging data of perforator vessels.

[0018] S2. Hemodynamic Modeling: Based on the image dataset obtained in step S1, a microcirculation model of the perforator flap is constructed using computational fluid dynamics methods to simulate the differences in blood perfusion under different transplant angles and vascular anastomosis methods, and output the flap blood perfusion efficiency score (0-100 points) and necrosis risk distribution map.

[0019] S3. AI Multi-Objective Optimization: Using the necrosis risk distribution map, negative tumor margin requirement, donor scar length <8cm, and breast aesthetic symmetry Δ volume ≤10% output from step S2 as constraints, the Pareto optimal solution is solved through an LSTM-DRL hybrid architecture to generate at least one transplantation plan that includes a perforator selection strategy and a volume compensation scheme. The scheme initializes weights through transfer learning and is fine-tuned with multi-center clinical data.

[0020] S4. Intraoperative Real-time Calibration: A time-synchronous detection mechanism is established using ICG fluorescence imaging equipment (detection frequency 3 frames / second) and OCT equipment (scanning resolution 10μm) to obtain real-time perforator vessel function data; when the blood flow velocity is detected to decrease by ≥30% compared with the preoperative predicted value, a level 3 warning is triggered, and the real-time data is fed back to the navigation system to dynamically update the surgical plan and AR navigation path, with an intraoperative decision delay of <5 seconds.

[0021] As a further improvement to the above technical solution:

[0022] The image processing module also includes an image noise reduction unit. The image noise reduction unit uses an adaptive Gaussian filtering algorithm to preprocess the CT thin-slice scan data to remove scanning noise. The consistency between the preprocessed image and the image manually processed by the expert is ≥98%, and the details of blood vessel branches with a diameter ≥0.3mm are preserved.

[0023] The AI ​​decision-making module also includes a built-in perforator vessel variation prediction submodule. Based on a database of 100,000 perforator anatomical atlases, the submodule uses a random forest algorithm to predict the probability of perforator vessel variation, with a variation prediction sensitivity of ≥93%. For high-probability variation cases, it generates alternative perforator selection schemes that include correction parameters for vessel diameter and course angle.

[0024] The augmented reality device of the dynamic navigation module also integrates eye-tracking function, which automatically optimizes the clarity and labeling priority of blood vessel projection by recognizing the surgeon's gaze focus, and activates the layered display mode when ≥3 important blood vessels appear in the field of vision at the same time.

[0025] In step S1, the interslice spacing and slice thickness of the CT scan data are both ≤0.625mm, and a dual-energy scanning mode is used to distinguish between blood vessels and glandular tissues, with a blood vessel contrast improvement of ≥40%.

[0026] In step S2, hemodynamic modeling also includes simulating the stress distribution of the blood vessel wall under different suture tensions, and automatically marking high-risk anastomosis areas when the predicted local stress exceeds the threshold of 3000Pa.

[0027] In step S4, the real-time calibration process also includes monitoring the oxygen saturation of the flap tissue using near-infrared spectroscopy. When the oxygen saturation of any region is ≤70% for more than 10 seconds, the navigation system automatically pushes suggestions for adjusting the vascular anastomosis angle.

[0028] It also includes a data storage and interaction module, which stores the patient's preoperative imaging data, intraoperative real-time monitoring data, AI decision-making schemes and postoperative follow-up data. It achieves data traceability through blockchain technology and supports seamless interaction with the hospital's HIS system and PACS system, forming a closed loop of data throughout the "preoperative-intraoperative-postoperative" cycle.

[0029] In step S3, the AI ​​multi-objective optimization also uses the risk of local recurrence 2 years after surgery as a constraint. By integrating clinical data such as tumor pathology type, grade, and immunohistochemical indicators, the postoperative local recurrence rate corresponding to the generated plan is ≤1.8%.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] 1. The image processing module of this invention uses "high-resolution CT (slice thickness ≤ 0.625 mm) + ultrasound contrast imaging multimodal fusion" technology, combined with an adaptive weighted dynamic fusion algorithm (dual constraints of grayscale value matching and vascular morphology features), to accurately correct the localization of micro-perforator vessels with a diameter < 0.5 mm, improving the accuracy of the perforator vessel tree model to ± 0.1 mm, which is 83% lower than the traditional solution (± 1.2 mm). At the same time, through adaptive Gaussian filtering preprocessing, the image structure similarity index (SSIM) is ≥ 0.95, and the consistency with manually processed images by experts is ≥ 98%, which fully meets the requirements of millimeter-level surgery for vascular localization and avoids temporary changes in surgical plans due to missed perforator detection.

[0032] 2. The AI ​​decision-making module of this invention adopts an LSTM-DRL hybrid architecture, breaking through the limitations of traditional static evaluation: On the one hand, the LSTM neural network can simulate the temporal changes in blood perfusion (pressure, flow rate, oxygen saturation) after flap transplantation, accurately predicting the flap survival rate 72 hours after surgery (AUC≥0.92), which is 23% more accurate than the traditional static evaluation (AUC=0.75); on the other hand, the DRL optimization unit constructs a three-dimensional reward function of "tumor safety-donor site damage-aesthetic effect", and solves the Pareto optimal solution through the improved NSGA-III algorithm, which can simultaneously meet the constraints of "negative surgical margin (tumor safety boundary ≥2mm), donor site scar <8cm, and breast Δ volume ≤10%), achieving a volume compensation rate ≥95%, and reducing the postoperative local recurrence rate to 1.8% (60% lower than the traditional procedure of 4.5%), fundamentally solving the limitations of a single decision objective.

[0033] 3. The dynamic navigation module of this invention constructs an active protection system through "bivariate Kalman filtering + graded force feedback": Bivariate Kalman filtering is based on the "image drift-tissue deformation" coupling model, which corrects AR navigation errors in real time (spatial positioning error ≤2mm) to ensure accurate projection of blood vessel course and flap cutting range; The graded force feedback device adopts a nonlinear resistance algorithm based on blood vessel diameter. When the distance between the instrument and the perforator blood vessel with a diameter <1mm is ≤2mm, the resistance increases exponentially with the decrease of distance, actively avoiding accidental damage to blood vessels; Combined with real-time patency detection of ICG fluorescence imaging (3 frames / second) and OCT (10μm resolution), the success rate of perforator blood vessel anastomosis is increased to ≥98% (15% higher than the traditional 85%), completely solving the problem of insufficient dynamics and safety of intraoperative navigation.

[0034] 4. This invention significantly improves the efficiency of the entire surgical process through automated and intelligent design: preoperative planning time is reduced from the traditional 120 minutes to 20 minutes, and intraoperative decision-making delay is less than 5 seconds, greatly reducing surgical interruptions; at the same time, the system has a built-in database of 100,000 perforator anatomy atlases, and through transfer learning and fine-tuning with multi-center clinical data, it can automatically generate standardized surgical plans (SSIM≥0.95), reducing reliance on surgeon experience and meeting the needs of promotion in primary hospitals; in addition, the data storage and interaction module uses blockchain technology to achieve full-cycle data traceability from "preoperative imaging to intraoperative monitoring to postoperative follow-up", and supports seamless integration with HIS / PACS systems, providing data support for technical standardization and clinical research.

[0035] 5. This invention significantly improves postoperative outcomes for patients through multi-dimensional technological synergy: In terms of complication control, the incidence of donor site complications has decreased from the traditional 22% to 3% (a reduction of 86%), avoiding increased trauma due to perforator variations; in terms of aesthetics, for breast-conserving surgery in the inner quadrant of breast cancer, the use of the lateral thoracic artery perforator flap (LTAP) combined with AI volume compensation reduces the postoperative upper pole breast retraction rate from 35% to 5%, and improves the patient's BREAST-Q score from 68 to 89, fully balancing the needs of "radical tumor removal" and "functional aesthetics," greatly improving the postoperative quality of life for patients, and conforming to the humanistic concept of modern tumor treatment.

[0036] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0037] Figure 1 This is the core flowchart of the system architecture;

[0038] Figure 2This is a flowchart of the method implementation steps. Detailed Implementation

[0039] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0040] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0042] I. Implementation Examples

[0043] This embodiment uses "inner quadrant breast cancer breast-conserving surgery + lateral thoracic artery perforator flap (LTAP) reconstruction" as the application scenario to verify the feasibility and superiority of the "intelligent navigation system and method for breast-conserving surgery perforator flap based on CT three-dimensional vascular reconstruction and AI dynamic blood flow analysis" described in this invention.

[0044] Experimental subject: One patient with inner quadrant breast cancer admitted to the breast surgery department of a tertiary hospital from January 2023 to June 2023 (age 45 years, BMI 22.3 kg / m², tumor diameter 2.1 cm, pathological type invasive ductal carcinoma, HER-2 negative, no lymph node metastasis) was selected. The preoperative assessment met the indications for breast-conserving surgery and "local extended tumor resection + LTAP perforator flap reconstruction" was planned.

[0045] Comparative approach: Traditional surgical procedure (thoracic dorsal artery perforator flap TDAP + manual experience planning), data from similar cases in the same period of 2022 at the same hospital (n=30).

[0046] Core equipment: The system of this invention includes a Siemens SOMATOMForce high-resolution CT (0.625mm slice thickness), a Philips EPIQ7C ultrasound angiography system, a HoloLens 3AR device, a Zeiss ICG fluorescence imaging system (detection frequency 3 frames / second), a Thorlabs OCT device (scanning resolution 10μm), customized graded force feedback surgical instruments, and an AI decision server equipped with an LSTM-DRL hybrid algorithm (CPU: Intel Xeon Gold 6348, GPU: NVIDIA A100).

[0047] II. Specific Implementation Steps

[0048] 1. Preoperative stage: Image acquisition and treatment planning (corresponding to S1-S3 of the amended claim 2)

[0049] 1.1 Preoperative image acquisition (S1)

[0050] CT scan: A Siemens SOMATOM ForceCT scanner was used, with a slice thickness of 0.625 mm and an inter-slice spacing of 0.625 mm. The scan area covered both breasts, chest wall, abdominal wall, and back (including the donor sites of the thoracodorsal artery, lateral thoracic artery, and inferior epigastric artery). A non-ionic contrast agent (iopromide 370 mgI / mL) was injected via the antecubital vein at a dose of 1.5 mL / kg and an injection rate of 3.5 mL / s. The "contrast agent peak triggering technology" (threshold CT value 180 HU) was enabled to capture the peak imaging period data of perforating vessels and obtain a DICOM format image dataset.

[0051] Supplemental ultrasound contrast imaging: A Philips EPIQ7C ultrasound system with a probe frequency of 9-12MHz was used. Ultrasound contrast agent (SonoVue) was injected percutaneously. The lateral thoracic artery region was scanned to obtain dynamic blood flow signals of small perforators with a diameter of <0.5mm (frame rate 20 frames / second), and the ultrasound contrast imaging sequence was output.

[0052] 1.2 Image processing and blood vessel modeling (corresponding to the image processing module of the improved claim 1)

[0053] Multimodal fusion correction: CT and ultrasound contrast imaging data are imported into the image processing module of this invention, and the "adaptive weighted dynamic fusion algorithm" is activated. Through gray value matching (CT vessel gray value threshold 150-250HU, ultrasound contrast imaging enhancement area gray value threshold 200-300HU) and vessel morphological features (diameter, course angle, branch density) dual constraints, the algorithm automatically corrects the problem of CT missing detection of small perforators with diameters of 0.3-0.5mm (the original CT missed two lateral thoracic artery perforators with a diameter of 0.4mm, which were accurately located after fusion). At the same time, "adaptive Gaussian filtering" preprocessing is used to reduce noise, and a three-dimensional model of the perforator vessel tree (generated by Mimics21.0 software) is output. The model accuracy is ±0.1mm, the structural similarity index (SSIM) is 0.97, and the consistency with expert manual annotation is 99%.

[0054] 1.3 Hemodynamic Modeling (S2)

[0055] Based on the above three-dimensional vascular model, an LTAP flap microcirculation model was constructed using ANSYS Fluent 2023R1 (Computational Fluid Dynamics CFD software). The simulation parameters included: flap transplantation angle (30°, 45°, 60°), vascular anastomosis method (end-to-end anastomosis, end-to-side anastomosis), and hemodynamic parameters (peak systolic velocity 18-25 cm / s, end-diastolic velocity 5-8 cm / s, oxygen saturation 95%-98%).

[0056] Output results: When the transplant angle is 45° and the end-to-end anastomosis is used, the flap blood flow perfusion efficiency score is 92 points (out of 100 points), and the necrosis risk area is only located at the edge of the flap (area <2%). This parameter is determined as the basic protocol.

[0057] 1.4 AI Multi-Objective Optimization (S3)

[0058] The constraints "tumor margin ≥2mm (pathologically confirmed negative margin), donor scar <6cm (actual design 5.8cm), breast Δ volume ≤8% (preoperative affected breast volume 280mL, healthy breast volume 300mL, compensated affected breast volume 295mL)" were input into the LSTM-DRL hybrid architecture of the AI ​​decision module.

[0059] LSTM layer: Based on the blood flow time series data of 100,000 perforator flaps, the predicted flap survival rate at 72 hours postoperatively was AUC=0.94, with blood flow pressure maintained at 80-100 mmHg (normal range 70-120 mmHg) and oxygen saturation ≥92%.

[0060] DRL layer: Construct a three-dimensional reward function of "tumor safety (weight 40%) - donor site damage (weight 30%) - aesthetic effect (weight 30%)", solve the Pareto optimal solution through the improved NSGA-III algorithm, and output the final solution: "LTAP perforator flap resection range 8cm×5cm, select 2 lateral thoracic artery perforators with diameter 0.6mm and 1 lateral thoracic artery with diameter 0.4mm, volume compensation rate 98%, and postoperative local recurrence risk prediction value 1.2%";

[0061] Meanwhile, the AI ​​perforator variation prediction submodule (random forest algorithm) predicts the patient's perforator variation probability at 8% (sensitivity 93%) based on the anatomical database and generates a backup plan: "If the LTAP perforator is abnormal, switch to the inferior epigastric artery perforator flap (DIEP), with a cut size of 10cm × 6cm."

[0062] The entire preoperative planning took 18 minutes (the traditional approach takes an average of 120 minutes).

[0063] 2. Intraoperative phase: Real-time navigation and calibration (corresponding to the dynamic navigation module of the improved claim 1 and S4 of claim 2)

[0064] 2.1 AR Navigation Initialization

[0065] The HoloLens 3AR device (65Hz refresh rate, 1.8mm spatial positioning error) is activated, and the perforator vessel path and flap cutting range output by AI are projected onto the surgical field. Through the "bivariate Kalman filter unit", intraoperative tissue traction (displacement ≤3mm) and respiratory motion (amplitude ≤2mm) data are collected in real time to establish an "image drift-tissue deformation" coupling model, dynamically correct navigation errors, and ensure that the deviation between the projection position and the actual blood vessel is ≤0.5mm.

[0066] 2.2 Graded force feedback protection

[0067] The surgeon used a custom-made force feedback surgical instrument to separate subcutaneous tissue. When the instrument was 2mm away from the perforating vessel, the device generated initial resistance (5N); when the distance decreased to 1mm, the resistance increased exponentially to 15N; when the distance was <0.5mm, the resistance reached 30N and triggered an audio-visual prompt, thus avoiding accidental injury to the perforating vessel throughout the procedure (the average accidental injury rate of the traditional method is 12%, while there was no accidental injury in this case).

[0068] 2.3 Real-time connectivity calibration

[0069] ICG fluorescence imaging: Perforator blood flow images were acquired every 3 seconds, showing a peak systolic velocity of 22 cm / s in the lateral thoracic artery perforator (preoperative prediction of 20 cm / s), which was in line with expectations;

[0070] OCT examination: Scanning the integrity of the perforator vessel wall shows that the vessel endothelium is smooth and the lumen is not narrowed (resolution 10μm, can identify damage of 0.01mm level).

[0071] At some point during the procedure, ICG testing revealed that the blood flow velocity of a 0.4mm diameter perforator had dropped to 15cm / s (a 25% decrease from the preoperative prediction, but below the 30% warning threshold). The system automatically pushed a suggestion to "fine-tune the anastomosis angle to 50°". After the adjustment, the flow velocity rose back to 18cm / s.

[0072] The entire intraoperative decision-making delay is ≤4 seconds (compared to an average of 12 seconds in the traditional approach).

[0073] 2.4 Flap transplantation and anastomosis

[0074] The LTAP flap was harvested according to the navigation range. Based on force feedback and AR guidance, the end-to-end anastomosis of three perforator vessels and the recipient vessel was completed (anastomosis time 25 minutes). The success rate of perforator vessel anastomosis was 100% (compared to an average of 85% in the traditional approach).

[0075] 3. Postoperative stage: Follow-up and data management

[0076] 3.1 Postoperative monitoring and outcomes

[0077] 24 hours post-surgery: The flap has good blood supply, a temperature of 36.5℃ (temperature difference from the healthy side ≤0.5℃), and an oxygen saturation of 93%.

[0078] 72 hours post-surgery: LSTM predicted survival rate consistent with actual survival (no necrotic areas).

[0079] Six months post-surgery: donor site scar length 5.8cm (<8cm), breast superior pole indentation rate 0 (average 35% in traditional TDAP protocol), breast Δ volume 7% (≤10%), patient BREAST-Q score 91 (average 68 in traditional protocol), and donor site complication rate (such as pain, numbness) 0 (average 22% in traditional protocol).

[0080] 12 months post-surgery: Follow-up CT scan showed that the perforating vessels were patent and there was no local recurrence (the average recurrence rate of the traditional treatment was 4.5%).

[0081] 3.2 Full lifecycle data management

[0082] The data storage and interaction module of this invention encrypts and stores "preoperative images (CT + ultrasound), intraoperative monitoring data (ICG + OCT + force feedback records), AI decision-making schemes, and postoperative follow-up results," and achieves data traceability through blockchain technology (unique identifier by hash value). It also seamlessly connects with the hospital's HIS system (patient basic information) and PACS system (image archiving) to form a complete surgical data closed loop, supporting subsequent clinical research and scheme iteration.

[0083] III. Verification of the Effects of the Examples

[0084]

[0085] IV. Conclusions of the Examples

[0086] 1. This embodiment verifies the feasibility of the system and method of the present invention through practical application in breast-conserving surgery for inner quadrant breast cancer:

[0087] Multimodal image fusion and LSTM-DRL algorithm enable precise localization of micro-perforators, dynamic prediction of blood flow, and optimization of multi-objective solutions, solving the pain points of traditional technology such as "missed detection, static assessment, and one-sided decision-making".

[0088] 2. Bivariate Kalman filtering and graded force feedback ensured intraoperative navigation accuracy and perforator safety, significantly improving the success rate of anastomosis.

[0089] 3. Improved efficiency throughout the entire process (preoperative planning, intraoperative decision-making) and standardized protocol generation to meet the needs of clinical promotion;

[0090] 4. Postoperative aesthetic results, complication control, and reduced risk of recurrence fully embody the three-in-one goal of breast preservation: "function-aesthetics-oncology," and have significant clinical value.

[0091] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A breast-conserving surgery perforator flap intelligent navigation system, characterized in that, include: The image processing module generates a three-dimensional model of perforating vessels with an accuracy of ±0.1mm based on high-resolution CT thin-slice scan data with a slice thickness ≤0.625mm using Mimics software. It also integrates a multimodal image fusion unit, which adopts an adaptive weighted dynamic fusion algorithm. Through grayscale value matching and vascular morphology features as dual constraints, it fuses CT image data and ultrasound contrast data to correct the localization of small perforating vessels with a diameter <0.5mm, and outputs a corrected perforating vessel tree model. The structural similarity index (SSIM) of the preprocessed image is ≥0.

95. The AI ​​decision-making module incorporates a hemodynamic prediction and optimization system based on an LSTM-DRL hybrid architecture. The LSTM neural network layer simulates the temporal changes in blood perfusion after perforator flap transplantation, including dynamic curves of blood pressure, flow velocity, and oxygen saturation, and predicts the flap survival rate (AUC≥0.92) 72 hours post-surgery. The deep reinforcement learning (DRL) optimization unit uses tumor invasion depth and HER-2 expression status as clinical feature parameters to construct a three-dimensional reward function that includes tumor safety boundaries (≥2mm), BREAST-Q aesthetic scores, and donor site damage. It solves for the Pareto optimal solution using an improved NSGA-III algorithm, outputting a personalized breast-conserving reconstruction plan with a volume compensation rate ≥95%. The dynamic navigation module includes an augmented reality device, a bivariate Kalman filter unit, and a graded force feedback device. The augmented reality device has a refresh rate of ≥60Hz and a spatial positioning error of ≤2mm, and is used to project the perforator vessel course and flap cutting range to the surgical field of view in real time. The bivariate Kalman filter unit performs dynamic error correction through an image drift-tissue deformation coupling model; the graded force feedback device is equipped with a nonlinear resistance algorithm based on the vessel diameter. When the distance between the surgical instrument and the perforating vessel with a diameter <1mm is ≤2mm, gradient resistance feedback is triggered. The resistance increases exponentially as the distance decreases, and the dynamic navigation module enables the perforating vessel anastomosis success rate to be ≥98%.

2. A method for planning a perforator flap in breast-conserving surgery, characterized in that, Includes the following steps: S1. Preoperative image acquisition: A high-resolution CT scanner with a slice thickness of ≤0.625mm was used to scan the patient's breast and flap donor area. The scanning range covered the bilateral breasts, chest wall, abdominal wall and back area to ensure that potential flap donor area vessels such as the thoracodorsal artery, lateral thoracic artery and inferior epigastric artery were included. At the same time, contrast agent peak triggering technology was used to capture the imaging data of perforator vessels. S2. Hemodynamic Modeling: Based on the image dataset obtained in step S1, a microcirculation model of the perforator flap is constructed using computational fluid dynamics methods to simulate the differences in blood perfusion under different transplant angles and vascular anastomosis methods, and output the flap blood perfusion efficiency score (0-100 points) and necrosis risk distribution map. S3. AI Multi-Objective Optimization: Using the necrosis risk distribution map, negative tumor margin requirement, donor scar length <8cm, and breast aesthetic symmetry Δ volume ≤10% output from step S2 as constraints, Pareto optimal solution is solved through LSTM-DRL hybrid architecture to generate at least one transplantation plan that includes perforator selection strategy and volume compensation scheme. The plan initializes weights through transfer learning and is fine-tuned with multi-center clinical data. S4. Intraoperative Real-time Calibration: A time-synchronous detection mechanism is established using ICG fluorescence imaging equipment (detection frequency 3 frames / second) and OCT equipment (scanning resolution 10μm) to obtain real-time perforator vessel function data; when the blood flow velocity is detected to decrease by ≥30% compared with the preoperative predicted value, a level 3 warning is triggered, and the real-time data is fed back to the navigation system to dynamically update the surgical plan and AR navigation path, with an intraoperative decision delay of <5 seconds.

3. The intelligent navigation system for perforator flaps in breast-conserving surgery according to claim 1, characterized in that, The image processing module also includes an image noise reduction unit. The image noise reduction unit uses an adaptive Gaussian filtering algorithm to preprocess the CT thin-slice scan data to remove scanning noise. The consistency between the preprocessed image and the image manually processed by the expert is ≥98%, and the details of blood vessel branches with a diameter ≥0.3mm are preserved.

4. The intelligent navigation system for perforator flaps in breast-conserving surgery according to claim 1, characterized in that, The AI ​​decision-making module also includes a built-in perforator vessel variation prediction submodule. Based on a database of 100,000 perforator anatomical atlases, the submodule uses a random forest algorithm to predict the probability of perforator vessel variation, with a variation prediction sensitivity of ≥93%. For high-probability variation cases, it generates alternative perforator selection schemes that include correction parameters for vessel diameter and course angle.

5. The intelligent navigation system for perforator flaps in breast-conserving surgery according to claim 1, characterized in that, The augmented reality device of the dynamic navigation module also integrates eye-tracking function, which automatically optimizes the clarity and labeling priority of blood vessel projection by recognizing the surgeon's gaze focus, and activates the layered display mode when ≥3 important blood vessels appear in the field of vision at the same time.

6. The method for planning perforator flaps in breast-conserving surgery according to claim 2, characterized in that, In step S1, the interslice spacing and slice thickness of the CT scan data are both ≤0.625mm, and a dual-energy scanning mode is used to distinguish between blood vessels and glandular tissues, with a blood vessel contrast improvement of ≥40%.

7. The method for planning perforator flaps in breast-conserving surgery according to claim 2, characterized in that, In step S2, hemodynamic modeling also includes simulating the stress distribution of the blood vessel wall under different suture tensions, and automatically marking high-risk anastomosis areas when the predicted local stress exceeds the threshold of 3000Pa.

8. The method for planning perforator flaps in breast-conserving surgery according to claim 2, characterized in that, In step S4, the real-time calibration process also includes monitoring the oxygen saturation of the flap tissue using near-infrared spectroscopy. When the oxygen saturation of any region is ≤70% for more than 10 seconds, the navigation system automatically pushes suggestions for adjusting the vascular anastomosis angle.

9. The intelligent navigation system for perforator flaps in breast-conserving surgery according to claim 1, characterized in that, It also includes a data storage and interaction module, which stores the patient's preoperative imaging data, intraoperative real-time monitoring data, AI decision-making schemes and postoperative follow-up data. It achieves data traceability through blockchain technology and supports seamless interaction with the hospital's HIS system and PACS system, forming a closed loop of data throughout the "preoperative-intraoperative-postoperative" cycle.

10. The method for planning perforator flaps in breast-conserving surgery according to claim 2, characterized in that, In step S3, the AI ​​multi-objective optimization also uses the risk of local recurrence 2 years after surgery as a constraint. By integrating clinical data such as tumor pathology type, grade, and immunohistochemical indicators, the postoperative local recurrence rate corresponding to the generated plan is ≤1.8%.

Images