Auxiliary drug delivery method, system and device

Abstract

The invention provides an auxiliary drug delivery method, system and device. The auxiliary drug delivery method comprises the following steps that S1, image data transmitted by DSA equipment in the TACE operation drug delivery stage process of a patient is obtained; s2, inputting the image data into the trained deep learning model; identifying a catheter head position in the image data by the deep learning model, generating a regurgitation rectangular region based on the catheter head position, performing pixel-level segmentation on the medicine in each frame of regurgitation rectangular region, and generating a segmentation mask; s3, multiple frames of data are obtained each time according to the time sequence of mask segmentation; and dividing into a discrimination period and a buffer period, and if the drug pixel data continuously appears in the discrimination period, determining that the reflux phenomenon exists, and reducing the injection flow rate by the injection device. According to the invention, the drug delivery strategy can be adjusted instead of manual work, and a technical foundation is laid for developing an intelligent and automatic TACE treatment system in the future. And meanwhile, the radiation exposure risk of medical staff is reduced, so that the treatment accuracy and safety are improved.

Description

Technical Field

[0001] This invention relates to the field of adjuvant drug delivery technology during transcatheter arterial chemoembolization (TACE), specifically to an adjuvant drug delivery method, system, and device, and particularly to a field of automated drug delivery rate adjustment during TACE based on deep learning. Background Technology

[0002] Hepatocellular carcinoma (HCC) is the major histological subtype of liver cancer, accounting for 90% of primary liver cancers and is the third leading cause of cancer-related mortality worldwide. Transcatheter arterial chemoembolization (TACE) is currently an important treatment for patients with early-stage and intermediate-to-late-stage HCC. Digital subtraction angiography (DSA) is an indispensable part of TACE treatment.

[0003] Currently, intraoperative drug delivery during TACE requires tailoring the dosage and rate based on the specific hemodynamics of the lesion to ensure that the drug effectively reaches the target tissue, maximizing treatment efficacy and minimizing collateral damage to normal tissues. This process typically relies on manual intervention by medical personnel, adjusting the rate based on factors such as tumor size, blood vessel diameter, and blood flow velocity. However, due to the complexity of the surgical environment and the precision requirements of the surgeon's operation, manually adjusting the drug delivery rate carries inherent errors and risks.

[0004] Although automated injection devices have emerged that use motors or other electrical equipment to adjust the injection flow rate after receiving a signal, and can be continuously adjusted, such as Chinese patents "A Micro-injection Pump (2017112305610)," "An Automatic Reversible Liquid Phase Injection Pump (2015107342216)," and "An Automatic Injection Pump (2012103709241)," it remains a challenge to provide a method that can continuously identify during surgery when the flow rate should be reduced to match the injection flow rate with the patient's blood flow status. Summary of the Invention

[0005] To address the aforementioned technical problems, an auxiliary drug delivery method and system are provided. This invention primarily utilizes the phenomenon of backflow occurring at the catheter tip when the flow rate is too high. A deep learning model is used to process the image data transmitted from the DSA device, dividing the backflow rectangular region. The drug within the backflow rectangular region is then segmented at the pixel level. Further verification is performed; if pixels containing drug continuously appear, it indicates that the drug injection flow rate is too high. This allows for a reduction in the drug injection rate, achieving automated drug delivery.

[0006] The technical means employed in this invention are as follows:

[0007] An adjunctive drug administration method includes the following steps:

[0008] S1: Acquire image data transmitted by the DSA device during the TACE drug administration phase;

[0009] S2: Input the image data into the trained deep learning model;

[0010] The deep learning model identifies the location of the catheter tip in the image data, generates a reflux rectangular region based on the location of the catheter tip, and performs pixel-level segmentation of the drug within the reflux rectangular region in each frame, generating a segmentation mask.

[0011] The catheter tip is the front end of the catheter used in TACE; the reflux rectangular area is a rectangular area with the catheter tip as the front edge, the two vessel walls on both sides as the left and right edges, and extending posteriorly to the catheter.

[0012] S3: Acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is the discrimination period. If at least N consecutive frames in the discrimination period identify drug pixel data in the segmentation mask, it is determined that there is a backflow phenomenon, and the injection device reduces the injection flow rate. The second group is the buffer period, and the data time in the transmission buffer period is adapted to the time of the injection device adjusting the flow rate. X≥10 frames, N≥3 frames, M≥N.

[0013] Preferably, X = 15 frames, N = 3 frames, M = 5 frames, and the flow rate adjustment time of the injection device is 0.7s.

[0014] Preferably, the deep learning model is the YOLOv8 deep learning model.

[0015] The present invention also discloses an auxiliary drug delivery system, comprising: an image data receiving module, a deep learning analysis module, a drug delivery control module, and a drug delivery module;

[0016] The image data receiving module is used to acquire image data transmitted by the DSA device during the TACE drug administration phase.

[0017] The deep learning analysis module is used to identify the location of the catheter tip in the image data, generate a reflux rectangular region based on the location of the catheter tip, perform pixel-level segmentation of the drug within the reflux rectangular region in each frame, and generate a segmentation mask; the catheter tip is the front end of the catheter during TACE; the reflux rectangular region is a rectangular region with the catheter tip as the front edge, the two sides of the vessel walls as the left and right edges, and extending to the rear of the catheter.

[0018] The drug delivery control module is used to acquire the segmentation mask and acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is the discrimination period. If at least N consecutive frames of data containing drug pixels in the segmentation mask are identified during the discrimination period, it is determined that there is a backflow phenomenon, and a signal to reduce the injection flow rate is provided to the drug delivery module; the second group is the buffer period, and the data time in the transmission buffer period is adapted to the flow rate adjustment time of the injection device; X≥10 frames, N≥3 frames, M≥N;

[0019] The drug delivery module is used to adjust the flow rate of the injection device of the drug delivery module according to the signal issued by the drug delivery control module.

[0020] The present invention also discloses an auxiliary drug delivery device, comprising a processor, a memory, and an injection device, wherein the processor executes a computer program stored in the memory to implement the above-mentioned auxiliary drug delivery method, and the injection device adjusts the injection flow rate according to the information issued by the processor.

[0021] When using this method, system, or apparatus, the flow rate of the injection device needs to be adjusted from high to low, taking full advantage of the phenomenon that backflow will occur when the flow rate is too high to adjust the flow rate of the injection device.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] 1. This invention utilizes images obtained from DSA equipment and processes them using a deep learning model. Based on the phenomenon that some medication may reflux when the flow rate is too high, a reflux rectangular region is first generated (this region does not necessarily contain refluxed medication; it is simply a region constructed using the learning results of the deep learning model). By setting the reflux rectangular region, the data segmentation range can be effectively reduced, interference from surrounding tissues and other irrelevant pixels can be avoided, and the data processing load can be reduced. Then, the reflux rectangular region is segmented pixel-level to generate a segmentation mask, thus highlighting the medication pixels within the reflux rectangular region. Afterwards, filtering is performed, with a discrimination period and a buffer zone within the filtering range. Multiple data points are filtered at once, increasing filtering efficiency. Simultaneously, continuous recognition is used to reduce the impact of single-frame errors (such as false positives) on subsequent flow rate adjustments. The buffer zone ensures the stability of feedback, avoiding frequent short-term adjustments, and provides sufficient time for the injection device to adjust the flow rate. If subsequent cycles still require flow rate adjustment commands, the infusion rate is further adjusted based on the previous infusion behavior to ensure precise control of the injection process.

[0024] 2. This invention eliminates the need for medical staff to manually adjust the flow rate of the injection device, especially since radiation may be generated during the procedure, thus better avoiding the exposure of medical staff to radiation.

[0025] 3. This invention adjusts the flow rate of the drug step by step from the high-level (high flow rate) setting of the injection device until backflow no longer occurs. This makes the drug flow rate match the patient's needs, allowing for fast and precise adjustment.

[0026] Based on the above reasons, this invention can be widely applied in fields such as drug delivery rate regulation. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of an auxiliary drug delivery method according to the present invention. Detailed Implementation

[0029] It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this invention can be combined with each other. The invention will now be described in detail with reference to the accompanying drawings and embodiments. To make the objectives, technical solutions, and advantages of the embodiments of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] Example 1

[0031] like Figure 1 As shown, the present invention provides an adjunctive drug administration method, comprising the following steps:

[0032] S1: Acquire image data transmitted by the DSA device during the TACE drug administration phase. This image data is captured by the DSA device and transmitted to the client via TCP for real-time processing. Each transmitted image frame is an 8-bit grayscale image (4 bytes high and 4 bytes wide) encoded in DICOM (.dcm) format. Upon receiving the image data, the client decodes the data. The received DICOM format data is read using a standard DICOM parser and converted into a two-dimensional matrix suitable for further processing. Subsequently, it is converted to PNG format using the OpenCV image library for storage, a format that supports lossless compression and fast access.

[0033] S2: Input the image data into the trained deep learning model;

[0034] The deep learning model identifies the location of the catheter tip in the image data, generates a reflux rectangular region based on the location of the catheter tip, and performs pixel-level segmentation of the drug within the reflux rectangular region in each frame, generating a segmentation mask.

[0035] The catheter tip is the front end of the catheter used in TACE; the reflux rectangular area is a rectangular area with the catheter tip as the front edge, the two vessel walls on both sides as the left and right edges, and extending posteriorly to the catheter.

[0036] S3: Acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is the discrimination period. If at least N consecutive frames in the discrimination period identify drug pixel data in the segmentation mask, it is determined that there is a backflow phenomenon, and the injection device reduces the injection flow rate. The second group is the buffer period, and the data time in the transmission buffer period is adapted to the time of the injection device adjusting the flow rate. X≥10 frames, N≥3 frames, M≥N.

[0037] In this embodiment, X = 15 frames, N = 3 frames, and M = 5 frames because experimental calculations showed that the injection device adjusts the flow rate in 0.7 seconds, and the buffer period of the last 10 frames provides 0.67 seconds (based on an image transmission rate of 15 frames / second). Although there is a slight difference (approximately 0.03 seconds) between this and the 0.7 seconds required by the injection pump, this difference has almost no impact on the overall stability of the system and the flow rate adjustment, and can be ignored. Therefore, the flow rate adjustment command can be implemented in a timely manner without significantly interfering with subsequent image processing. X = 15 frames is based on the characteristic that the acquired dynamic angiography data is 15 frames (this 15-frame interval is common; currently, mainstream companies like Philips, Siemens, and domestic companies like Neusoft use 15-frame acquisition). A cycle with a period of 15 frames was designed, with the first 5 frames serving as the discrimination period and the last 10 frames as the buffer period.

[0038] In this embodiment, the deep learning model used is the YOLOv8 deep learning model.

[0039] YOLOv8, a deep learning model, is a common model widely used in fields such as medical image segmentation. This model consists of a Backbone, Neck, and Head section. The Backbone uses C2f modules as its basic unit, comprising 5 ConvModules, 4 C2f modules, and 1 SPPF module. The Neck section is responsible for multi-scale feature fusion, fusing feature maps from different stages of the Backbone. The Neck section includes SPPF, PAA, and PAN modules. The Head section further processes the features output from the Neck section to generate the final detection result. The Head section contains three detect blocks, each containing two branches. Each branch consists of two ConvModules and one separate Conv2d convolutional layer. For pixel-level segmentation, replacing the detect blocks in the Head section with segment blocks generates a segmentation mask.

[0040] Example 2

[0041] The present invention also discloses an auxiliary drug delivery system, comprising: an image data receiving module, a deep learning analysis module, a drug delivery control module, and a drug delivery module;

[0042] The image data receiving module is used to acquire image data transmitted by the DSA device during the TACE drug administration phase.

[0043] The deep learning analysis module is used to identify the location of the catheter tip in the image data, generate a reflux rectangular region based on the location of the catheter tip, perform pixel-level segmentation of the drug within the reflux rectangular region in each frame, and generate a segmentation mask; the catheter tip is the front end of the catheter during TACE; the reflux rectangular region is a rectangular region with the catheter tip as the front edge, the two sides of the vessel walls as the left and right edges, and extending to the rear of the catheter.

[0044] The drug delivery control module is used to acquire the segmentation mask and acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is a discrimination period. If at least N consecutive frames of data containing drug pixels in the segmentation mask are identified during the discrimination period, it is determined that there is a backflow phenomenon, and a signal to reduce the injection flow rate is provided to the drug delivery module. The second group is a buffer period, and the data time in the transmission buffer period is adapted to the flow rate adjustment time of the injection device. X≥10 frames, N≥3 frames, M≥N; preferably X=15 frames, N=3 frames, M=5 frames.

[0045] The drug delivery module is used to adjust the flow rate of the injection device of the drug delivery module according to the signal issued by the drug delivery control module.

[0046] Example 3

[0047] The present invention also discloses an auxiliary drug delivery device, including a processor, a memory, and an injection device, wherein the processor executes a computer program stored in the memory to implement an auxiliary drug delivery method mentioned in Embodiment 1, and the injection device adjusts the injection flow rate according to the information issued by the processor.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adjunctive drug delivery method, characterized in that, Includes the following steps: S1: Acquire image data transmitted by the DSA device during the TACE drug administration phase; S2: Input the image data into the trained deep learning model; The deep learning model identifies the location of the catheter tip in the image data, generates a reflux rectangular region based on the location of the catheter tip, and performs pixel-level segmentation of the drug within the reflux rectangular region in each frame, generating a segmentation mask. The catheter tip is the front end of the catheter used in TACE; the reflux rectangular area is a rectangular area with the catheter tip as the front edge, the two vessel walls on both sides as the left and right edges, and extending posteriorly to the catheter. S3: Acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is the discrimination period. If at least N consecutive frames in the discrimination period identify drug pixel data in the segmentation mask, it is determined that there is a backflow phenomenon, and the injection device reduces the injection flow rate. The second group is the buffer period, and the data time in the transmission buffer period is adapted to the time of the injection device adjusting the flow rate. X≥10 frames, N≥3 frames, M≥N.

2. The adjunctive drug delivery method according to claim 1, characterized in that, X = 15 frames, N = 3 frames, M = 5 frames, and the injection device adjusts the flow rate in 0.7s.

3. The adjunctive drug delivery method according to claim 1, characterized in that, The deep learning model used is YOLOv8.

4. An auxiliary drug delivery system, characterized in that, include: Image data receiving module, deep learning analysis module, drug delivery control module, and drug delivery module; The image data receiving module is used to acquire image data transmitted by the DSA device during the TACE drug administration phase. The deep learning analysis module is used to identify the location of the catheter tip in the image data, generate a reflux rectangular region based on the location of the catheter tip, perform pixel-level segmentation of the drug within the reflux rectangular region in each frame, and generate a segmentation mask; the catheter tip is the front end of the catheter during TACE; the reflux rectangular region is a rectangular region with the catheter tip as the front edge, the two sides of the vessel walls as the left and right edges, and extending to the rear of the catheter. The drug delivery control module is used to acquire the segmentation mask and acquire X frames of data each time according to the time sequence of the segmentation mask; and divide the X frames of data into two groups. The data in the first group is no less than M frames, which is the discrimination period. If at least N consecutive frames of data containing drug pixels in the segmentation mask are identified during the discrimination period, it is determined that there is a backflow phenomenon, and a signal to reduce the injection flow rate is provided to the drug delivery module; the second group is the buffer period, and the data time in the transmission buffer period is adapted to the flow rate adjustment time of the injection device; X≥10 frames, N≥3 frames, M≥N; The drug delivery module is used to adjust the flow rate of the injection device of the drug delivery module according to the signal issued by the drug delivery control module.

5. An auxiliary drug delivery device, characterized in that, The device includes a processor, a memory, and an injection device, wherein the processor executes a computer program stored in the memory to implement an auxiliary drug delivery method as described in any one of claims 1-3, and the injection device adjusts the injection flow rate according to information issued by the processor.

Images