High-precision detection method for abnormal plasma cells
Through the intelligent AI recognition model combined with FISH technology, the bone marrow fluid treatment and slide preparation steps are optimized, which solves the problem of inaccurate detection in multiple myeloma diagnosis, and achieves high-precision and low-cost abnormal plasma cell detection, which is suitable for efficient diagnosis of multiple myeloma.
Patent Information
- Application Number
- CN202510436140.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-09
- Publication Date
- 2025-07-04
AI Technical Summary
In the prior art, the diagnosis of multiple myeloma depends on fluorescence in situ hybridization technology (FISH), but conventional FISH technology destroys cell structure and cannot accurately locate myeloma cells. Moreover, the CD138 magnetic bead sorting method is costly and the sorting efficiency is low, resulting in inaccurate detection results and affecting the treatment effect.
The intelligent AI recognition model is used in combination with fluorescence in situ hybridization technology (FISH), and optimizes bone marrow fluid treatment and slide preparation steps, and combines a trained AI model to identify fluorescent signals of abnormal plasma cells, reducing artificial errors, improving detection accuracy and efficiency, and reducing costs.
It realizes high-precision detection of abnormal plasma cells, reduces detection costs, reduces artificial errors, and improves the accuracy and consistency of detection results. It is suitable for applications between different laboratories.
Smart Images
Figure FT_1 
Figure FT_2
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of medical detection, and particularly relates to a highly accurate detection method for abnormal plasma cells. Background Art
[0002] Fluorescence in situ hybridization (FISH) is a molecular cytogenetic technique that uses fluorescently labeled DNA probes to hybridize with specific DNA sequences in cells or tissue sections, enabling the visualization of the location of specific genes or chromosomal regions under a fluorescence microscope. This technique is widely used in fields such as gene mapping, chromosomal abnormality detection, and cancer research. Through FISH, researchers can directly observe the distribution of genes within the cell nucleus, providing important molecular biological evidence for disease diagnosis and treatment.
[0003] Multiple myeloma (MM) is a malignant tumor originating from bone marrow plasma cells, characterized by the formation of tumors by abnormally proliferating plasma cells in the bone marrow, leading to bone destruction and bone pain. This disease usually affects middle-aged and elderly populations, and its exact cause is not fully understood, but it may be related to genetic, environmental factors, and immune system abnormalities. The diagnosis of multiple myeloma relies on blood and urine tests, bone marrow biopsy, and imaging examinations. Treatment strategies include chemotherapy, targeted therapy, immunomodulatory therapy, and hematopoietic stem cell transplantation, etc. In both the Mayo Clinic guidelines and the NCCN guidelines, the prognosis of multiple myeloma is stratified based on the results of fluorescence in situ hybridization tests to assist clinical medication guidance.
[0004] Conventional FISH techniques destroy the cell structure and cannot accurately locate myeloma cells; in patients with multiple myeloma (MM), the proportion of myeloma cells is often low, and the detection results of conventional FISH are not precise enough, hindering the risk stratification of MM patients and affecting the treatment effect. The cells obtained by CD138 magnetic bead sorting contain some normal plasma cells, and the CD138 magnetic bead sorting method is affected by the sorting efficiency, and the purity obtained after sorting is often low, resulting in inaccurate FISH results, and the cost of CD138 magnetic bead sorting is relatively high. Therefore, there is a need for a simple, rapid, low-cost, and highly accurate detection method for abnormal plasma cells. Summary of the Invention
[0005] The object of the present invention is to provide a highly accurate detection method for abnormal plasma cells in view of the deficiencies of the prior art, which can visually observe the FISH signals in abnormal plasma cells and has strong specificity.
[0006] To achieve the above object, the present invention adopts the following technical solutions: A highly accurate detection method for abnormal plasma cells, comprising the following steps:
[0007] (1) Immediately mix the extracted bone marrow fluid with an anticoagulant, and successively perform dilution and centrifugation treatments. Collect the white blood cell layer. After adjusting the cell concentration, add the cell suspension to the funnel of a cytocentrifuge smear maker to make slides.
[0008] (2) Place the slides into a pre-treated 90% formamide / 2×SSC solution to permeate the cell membranes, and then soak the slides in gradient alcohol. Take out the slides and dry them.
[0009] (3) Add a probe to the slides, cover with a coverslip, seal and dry, and then place in a hybridization instrument for denaturation hybridization.
[0010] (4) Remove the coverslip from the hybridized slides for washing. Then add an anti-fading blocking agent to the slides and cover with a coverslip again to cover the slides.
[0011] (5) Scan the whole slide with a trained intelligent AI recognition model to generate a fluorescence signal picture. Then remove the coverslip, stain the slides with Wright-Giemsa stain, and scan the whole slide again with the intelligent AI recognition model. Combine the scanning results with the fluorescence signal picture to identify immature plasma cells and observe the fluorescence signals in the immature plasma cells.
[0012] Further, the anticoagulant in step (1) is sodium heparin or EDTA.
[0013] Further, in step (1), the dilution of the bone marrow is to dilute the bone marrow with RPMI 1640 / 5% FBS, and the dilution volume ratio is 1:1 - 5. Place the diluted cells into lymphocyte separation medium Ficoll. The conditions for centrifugation are either at room temperature, 600 - 800×g or 1800 - 2000 revolutions per minute for centrifugation for 25 minutes. Adjust the cell concentration to 1×10 6 / mL.
[0014] Further, the 90% formamide / 2×SSC solution in step (2) is to mix 100% formamide and 20×SSC in a volume ratio of 9:1, and adjust the pH of the solution to 7.0. The pretreatment is to preheat the 90% formamide / 2×SSC in a 37°C water bath.
[0015] Further, soaking the slides in gradient alcohol in step (2) is to successively soak the slides in pre-cooled 70% alcohol at -20°C for 2 minutes, pre-cooled 85% alcohol at -20°C for 2 minutes, and pre-cooled 100% alcohol at -20°C for 2 minutes.
[0016] Further, in step (3), 10 μL of the probe is added for each cell spot; for the denaturation and hybridization, the glass slide is placed in a hybridization instrument for denaturation at 75 °C for 15 min, and then hybridization is carried out at 42 °C for at least 15 h.
[0017] Further, in step (4), for the slide washing, the glass slide is placed in a 50% formamide / 2×SSC solution at 55 °C for 30 min, then the glass slide is placed in a 2×SSC solution at 37 °C for 8 min, and then washed with 1×PBD at room temperature for 3 min;
[0018] The 50% formamide / 2×SSC solution is a 50% formamide / 2×SSC solution, which is mixed with 100% formamide, 20×SSC, and distilled water in a volume ratio of 5:1:4, and the pH value is adjusted to 7.0;
[0019] The 2×SSC solution is mixed with 20×SSC and distilled water in a volume ratio of 1:9, and the pH value is adjusted to 7.0;
[0020] The 1×PBD is prepared by adding 10 mL of NP-40 to 1000 mL of 10×PBS solution, mixing well to prepare a 10×PBD solution, and then diluting the 10×PBD with distilled water in a volume ratio of 9:1, and adjusting the pH value to 7.0.
[0021] Further, in step (4), 10 μL of the anti-attenuation blocking agent is added for each cell spot for the addition of the blocking agent.
[0022] Further, the intelligent AI recognition model in step (5) is trained using the following steps:
[0023] S1. Upload the observation and recognition records of the cell fluorescence signals as the original learning database for training and constructing the model;
[0024] S2. Select a convolutional neural network, the Resnet-50 model as the deep learning architecture according to the characteristics of the fluorescence signals to train the basic model; for the establishment of the sample library, the primary and immature plasma cells of multiple myeloma after CD138 magnetic bead sorting are used to establish the sample library for basic model training.
[0025] S3. Optimize and improve the performance of the basic model, and use methods such as adjusting the model architecture, improving the loss function, using regularization techniques to reduce overfitting, adjusting the learning rate and batch size for optimization and improvement;
[0026] S4. Test the model to improve the running performance of the model;
[0027] S5. Complete the construction and creation of the intelligent AI recognition model.
[0028] Further, for Wright-Giemsa staining of the glass slide in step (5), a Wright-Giemsa staining solution and a buffer solution with a volume ratio of 1:1 are added, stained at room temperature for 13 min, and then dried.
[0029] The present invention provides a highly accurate detection method for abnormal plasma cells, having the following beneficial effects:
[0030] 1. By combining an intelligent AI recognition model and fluorescence in situ hybridization technology (FISH), the present invention improves the accuracy and efficiency of abnormal plasma cell detection and reduces costs. The application of the intelligent AI recognition model can avoid using CD138 magnetic beads for sorting, thus significantly reducing reagent costs and being able to reduce human errors, improve the recognition ability of fluorescence signals, and thereby enhance the accuracy of detection results.
[0031] 2. The method of the present invention details the whole process from bone marrow fluid extraction, dilution, centrifugation, cell preparation to FISH detection, including the specific operations of the preservation treatment and washing steps of the glass slide. The optimization of these steps helps to improve the stability and repeatability of the sample and ensure the reliability of the detection results. Through a standardized operation process, this method helps to promote the application among different laboratories and achieve the consistency of results. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a flowchart of the highly accurate detection method for abnormal plasma cells of the present invention.
[0033] Figure 2 It is a FISH image of a bone marrow patient in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0035] Embodiment:
[0036] Please refer to the attached Figure 1 , the embodiment of the present invention provides a highly accurate detection method for abnormal plasma cells, including the following steps:
[0037] Step 1. Prepare the slides. The extracted bone marrow fluid should be immediately mixed with an anticoagulant. Any one of sodium heparin or EDTA can be used as the anticoagulant. Dilute and centrifuge the bone marrow successively, then collect the white blood cell layer, wash it with RPMI 1640 / 5% FBS at least once, then count the cells, adjust the cell concentration to 1×106 / ml, and then add 100 ul of the cell suspension into the funnel of a cytocentrifuge. Centrifuge at 1000 revolutions per minute at room temperature for 5 minutes. After completion, preserve the slide with the cell suspension. Among them, the specific steps for successively diluting and centrifuging the bone marrow are as follows:
[0038] S1. Dilute the bone marrow with RPMI 1640 / 5% FBS at a dilution ratio of either 1:1 or 1:5, and carefully place the diluted cells into Ficoll.
[0039] S2. Under room temperature conditions, centrifuge at either 600 - 800×g or 1800 - 2000 revolutions per minute for 25 minutes.
[0040] Among them, the specific steps for preserving the slide are as follows:
[0041] S1. Place the slide in a slide staining jar containing 100% alcohol for 5 - 10 minutes for fixation.
[0042] S2. The fixed slide needs to be baked in an oven at 37℃ overnight.
[0043] S3. Preserve the slide in a -20℃ refrigerator to avoid repeated freezing and thawing.
[0044] S4. Use a diamond pen to draw a circle on the back of the slide to mark the cell position for subsequent operations.
[0045] Step 2. Prepare the cells for FISH. First, prepare and pre-treat the 90% formamide / 2×SSC solution, and prepare 70%, 85%, and 100% alcohol pre-cooled at -20℃. Place the slide in the 90% formamide / 2×SSC solution at 37℃ for 5 minutes to penetrate the cell membrane, then place the slide in the pre-cooled 70% alcohol for 2 minutes, 85% alcohol for 2 minutes, and 100% alcohol for 2 minutes. Take out the slide and stand it on a paper towel to dry the slide. Among them, the specific steps for preparing and pre-treating the 90% formamide / 2×SSC solution are as follows:
[0046] S1. Mix 100% formamide and 20×SSC in a ratio of 9:1, and adjust the pH of the solution to 7.0 to obtain the 90% formamide / 2×SSC solution.
[0047] S2. Preheat the 90% formamide / 2×SSC in a 37℃ water bath.
[0048] Step 3: Add probes and seal the glass slide. Provide 10 μl of probes for each cell spot, cover the cell spot with a coverslip, apply the adhesive around the glass slide, and place it in the hybridization instrument after it dries.
[0049] Step 4: Hybridization. Place the glass slide in the hybridization instrument for denaturation at 75 °C for 15 minutes, and then perform hybridization at 42 °C for at least 15 hours.
[0050] Step 5: Wash the slide. Prepare 50% formamide / 2×SSC solution, 2×SSC solution, and PBD solution. Remove the adhesive, uncover the coverslip, place the glass slide in the 50% formamide / 2×SSC solution at 55 °C for 30 minutes, then place the glass slide in the 2×SSC solution at 37 °C for 8 minutes, and then wash it with 1×PBD at room temperature for 3 minutes. Finally, add 10 μl of anti-fading blocker for each cell spot and cover the glass slide with a rectangular coverslip with a specification of 20×60 cm. The specific steps for preparing the 50% formamide / 2×SSC solution, 2×SSC solution, and PBD solution are as follows:
[0051] Prepare 50% formamide / 2×SSC solution. Mix 100% formamide, 20×SSC, and distilled water in a ratio of 5:1:4, adjust the pH value to 7.0, and preheat it to 55 °C in a water bath.
[0052] Prepare 2×SSC solution. Mix 20×SSC and distilled water in a ratio of 1:9, adjust the pH value to 7.0, and preheat it to 37 °C in a water bath.
[0053] Prepare PBD solution. Add 10 ml of NP-40 to 1000 ml of 10×PBS solution, mix well to prepare 10×PBD solution, and then dilute 10×PBD with distilled water in a ratio of 9:1 and adjust the pH value to 7.0 (1×PBD).
[0054] Step 6: Scan the slide. Scan the whole slide with a trained intelligent AI recognition model to observe the fluorescence signal in the cells.
[0055] Step 7: Wright-Giemsa staining. Uncover the coverslip, add 13 drops of Wright-Giemsa stain and buffer respectively, stain at room temperature for 13 minutes, and wait for it to dry.
[0056] Step 8: Scan the slide again. Scan the whole slide with the trained intelligent AI recognition model, synthesize the scanning result with the fluorescence signal picture, identify immature plasma cells through AI, and observe the fluorescence signal in the immature plasma cells.
[0057] Among them, the training steps of the intelligent AI recognition model include:
[0058] S1. Upload the observed and recognized records of cell fluorescence signals as the original learning database for training and constructing the model.
[0059] S2. Select the convolutional neural network (CNN), Resnet-50 model as the deep learning architecture according to the characteristics of the fluorescence signals to train the basic model; establish a sample library using the primitive and immature plasma cells of multiple myeloma sorted by CD138 magnetic beads for basic model training.
[0060] S3. Optimize and improve the performance of the basic model by adjusting the model architecture, improving the loss function, using regularization techniques to reduce overfitting, adjusting the learning rate and batch size, etc.
[0061] (1) Transfer learning and application of pre-trained weights
[0062] Based on the pre-trained weights of ResNet-50, adapt to the feature distribution of myeloma cell images by freezing the underlying convolutional layers (retaining the general feature extraction ability) and fine-tuning the top network.
[0063] Add a custom layer to the model head:
[0064] Global Average Pooling layer: Replace the fully connected layer to reduce the number of parameters and prevent overfitting.
[0065] Attention mechanism module: Introduce the channel attention SE-Net to enhance the model's focusing ability on key regions such as nuclear morphology and fluorescence signal intensity.
[0066] (2) Regularization and optimization of training strategies
[0067] Morphological enhancement: Random rotation (±30°), translation (±10%), scaling (0.8 - 1.2 times), elastic deformation to simulate cell deformation.
[0068] Fluorescence signal simulation: Add Gaussian noise, adjust contrast / brightness to simulate image variations under different staining conditions.
[0069] S4. Test the model to improve the model's running performance.
[0070] In this stage, verify the clinical practicability and robustness of the model through multi-dimensional tests, which specifically include the following steps:
[0071] 1. Cross-validation and index quantification
[0072] Randomly divide the sample library into 5 subsets, and train with 4 subsets and test with 1 subset in turn to ensure the statistical significance of the model performance evaluation.
[0073] 2. Adversarial Testing and Robustness Verification
[0074] Noise Interference Test
[0075] Add Gaussian noise, motion blur, and occlusion of different intensities to the test images to evaluate the recognition stability of the model under low-quality images.
[0076] 3. Clinical Comparative Trials
[0077] Comparison with Manual Diagnosis: Invite 3 senior pathologists to perform blind annotation on the same batch of samples, calculate the consistency between the model and the physicians, and the Kappa value ≥ 0.85.
[0078] S5. Complete the construction and creation of the intelligent AI recognition model.
[0079] Figure 2 This is the FISH image of a myeloma patient. The left side is the scanned image under fluorescence. Under fluorescence, cells cannot be distinguished by morphology. The common method is to sort cells to increase the proportion of abnormal plasma cells, so as to obtain the true FISH result of plasma cells without being interfered by other cells. The right figure is the scanned image under the oil immersion lens. The plasma cells are identified by artificial intelligence, and then the scanning position is located, so that the signal of this plasma cell is one green and two fusions, which is positive. This greatly saves the specimens and does not require sorting beads, saving costs.
[0080] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A highly accurate detection method for abnormal plasma cells, characterized in that It includes the following steps: (1) Immediately mix the extracted bone marrow fluid with an anticoagulant, sequentially perform dilution and centrifugation treatments, collect the white blood cell layer, adjust the cell concentration, and then add the cell suspension into the funnel of a cytocentrifuge smear maker to make slides; (2) Place the slides into a pre-treated 90% formamide / 2×SSC solution to infiltrate the cell membrane, then soak the slides in gradient alcohol, and take out the slides to dry; (3) Add a probe to the slides, cover with a coverslip, seal and dry, and then place them in a hybridization instrument for denaturing hybridization; (4) Remove the coverslip from the hybridized slides for washing, then add an anti-fading blocker to the slides, and cover with a coverslip again to cover the slides; (5) Scan the whole slide with a trained intelligent AI recognition model to generate a fluorescence signal picture; then remove the coverslip, stain the slides with Wright-Giemsa stain, and scan the whole slide again with the intelligent AI recognition model. Combine the scanning results with the fluorescence signal picture to identify immature plasma cells and observe the fluorescence signals in the immature plasma cells.
2. The highly accurate detection method for abnormal plasma cells according to claim 1, wherein: The anticoagulant in step (1) is sodium heparin or EDTA.
3. The highly accurate detection method for abnormal plasma cells according to claim 1, characterized in that: The dilution of the bone marrow in step (1) is to dilute the bone marrow with RPMI 1640 / 5% FBS, with a dilution volume ratio of 1:1 to 5, and place the diluted cells into lymphocyte separation medium Ficoll; the conditions for centrifugation are any one of the following two schemes: under room temperature conditions, centrifuge at 600 - 800×g or 1800 - 2000 revolutions per minute for 25 minutes; adjust the cell concentration to 1×10 6 / mL.
4. The highly accurate detection method for abnormal plasma cells according to claim 1, characterized in that: The 90% formamide / 2×SSC solution in step (2) is prepared by mixing 100% formamide and 20×SSC in a volume ratio of 9:1, and adjusting the pH of the solution to 7.
0. The pre-treatment is to preheat the 90% formamide / 2×SSC in a 37°C water bath.
5. The high-precision detection method for abnormal plasma cells according to claim 1, characterized in that: The soaking of the slides in gradient alcohol in step (2) is to sequentially soak the slides in pre-cooled 70% alcohol at -20°C for 2 min, pre-cooled 85% alcohol at -20°C for 2 min, and pre-cooled 100% alcohol at -20°C for 2 min.
6. The high-precision detection method for abnormal plasma cells according to claim 1, characterized in that: The addition of the probe in step (3) provides 10 μL of probe for each cell spot; the denaturing hybridization is to place the slides in a hybridization instrument for denaturation at 75°C for 15 min, and then perform hybridization at 42°C for at least 15 h.
7. The highly precise detection method for abnormal plasma cells according to claim 1, characterized in that: The washing of the slides in step (4) is to place the slides in a 50% formamide / 2×SSC solution at 55°C for 30 min, then place the slides in a 2×SSC solution at 37°C for 8 min, and then wash with 1×PBD at room temperature for 3 min; Among them, the 50% formamide / 2×SSC solution is a 50% formamide / 2×SSC solution, which is mixed with 100% formamide, 20×SSC, and distilled water in a volume ratio of 5:1:4, and the pH value is adjusted to 7.0; The 2×SSC solution is mixed with 20×SSC and distilled water in a volume ratio of 1:9, and the pH value is adjusted to 7.0; The 1×PBD is prepared by adding 10 mL of NP-40 to 1000 mL of 10×PBS solution, mixing well to prepare a 10×PBD solution, and then diluting the 10×PBD with distilled water in a volume ratio of 9:1, and adjusting the pH value to 7.
0.
8. The highly accurate detection method for abnormal plasma cells according to claim 1, characterized in that: The addition of the blocker in step (4) is to add 10 μL of anti-fading blocker for each cell spot.
9. The highly accurate detection method for abnormal plasma cells according to claim 1, wherein: The intelligent AI recognition model in step (5) is trained using the following steps: S1. Upload the observation and recognition records of cell fluorescence signals as the original learning database for training and constructing the model; S2. Select the convolutional neural network and Resnet-50 model according to the characteristics of the fluorescence signals as the deep learning architecture to train the basic model; establish a sample library by using the primary and immature plasma cells of multiple myeloma sorted by CD138 magnetic beads to train the basic model; S3. Optimize and improve the performance of the basic model, and adjust the model architecture, improve the loss function, use regularization techniques to reduce overfitting, and adjust the learning rate and batch size for optimization and improvement methods; S4. Test the model to improve the running performance of the model; S5. Complete the construction and creation of the intelligent AI recognition model.
10. The high-precision detection method for abnormal plasma cells according to claim 1, characterized in that: The Wright-Giemsa staining of the glass slide in step (5) is to add Wright-Giemsa staining solution and buffer with a volume ratio of 1:1, stain at room temperature for 13 minutes, and dry.