Medical image segmentation system and method based on interactive point selection and adaptive ROI regulation

Abstract

The application discloses a medical image segmentation system and method based on interactive point selection and adaptive ROI regulation, which comprises a data input module, a parameter persistence module, an interactive point selection module, an adaptive segmentation module, a three-dimensional interpolation module and a visualization export module, and relates to the technical field of medical image processing. The medical image segmentation system and method based on interactive point selection and adaptive ROI regulation automatically completes ROI extraction and threshold calculation through single-point clicking of a user, dynamically determines a gray threshold value by using a statistical model and performs morphological optimization, realizes high-precision segmentation of two-dimensional and three-dimensional images, supports multiple formats of medical image input, realizes interactive segmentation, parameter persistence and result visualization export by using a PyQt interface, and has the advantages of strong adaptability, high robustness and no need of training samples compared with a traditional threshold method and a deep learning model, and is suitable for scientific research and clinical auxiliary diagnosis of medical images such as MRI and CT.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a medical image segmentation system and method based on interactive point selection and adaptive ROI control. Background Technology

[0002] In modern medical image analysis, accurate segmentation of lesion areas is an important prerequisite for volume measurement, radiotherapy planning, preoperative assessment, intraoperative guidance, and efficacy evaluation. Current segmentation methods include threshold-based, region growing, and deep learning networks (such as UNet).

[0003] The reference patent title is: A cross-modal medical image segmentation method, system and device (Patent Publication No.: CN117437420A, Patent Publication Date: 2024-01-23), which includes: constructing a cross-domain generator based on CUT and CycleGan and training it to obtain a trained cross-domain generation model; inputting an MRI image into the trained cross-domain generation model to generate a pseudo-CT image; training an image segmentation model based on the pseudo-CT image to obtain a trained image segmentation model; inputting the image to be segmented into the trained image model for segmentation to obtain segmented CT labels, which can perform accurate image segmentation with limited label data.

[0004] Based on the description in the above documents, existing segmentation methods include thresholding, region growing, and deep learning networks (such as UNet). Thresholding is simple and fast, but it is sensitive to grayscale distribution; region growing depends on the initial seed point and is prone to overflow; while deep learning methods require a large amount of labeled data and have high training costs. In addition, existing systems generally suffer from problems such as the separation of two-dimensional and three-dimensional segmentation processes, non-reusable parameters, and unstable results. To address these issues, this invention provides a medical image segmentation system and method based on interactive point selection and adaptive ROI control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a medical image segmentation system and method based on interactive point selection and adaptive ROI control. This solves the problems of existing segmentation methods, including those based on thresholding, region growing, and deep learning networks (such as UNet). Thresholding is simple and fast, but sensitive to grayscale distribution; region growing depends on initial seed points and is prone to overflow; while deep learning methods require a large amount of labeled data, resulting in high training costs. In addition, existing systems generally suffer from problems such as the separation of two-dimensional and three-dimensional segmentation processes, non-reusable parameters, and unstable results.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a medical image segmentation system based on interactive point selection and adaptive ROI control, comprising:

[0007] The data input module is used to read medical images in various image formats, automatically parse pixel spacing, layer spacing, and window width and window level information, and normalize grayscale by default.

[0008] The parameter persistence module uses QSettings to save the adaptive threshold coefficient K_STD, the lowest percentile elimination ratio, the ROI start, the maximum size and step size, the connected component area and the slenderness ratio threshold, the drawing style and the export path segmentation and display parameters across sessions, and automatically loads them when the segmentation system starts.

[0009] The interactive point selection module, based on the PyQt graphics event mechanism, obtains seed points in two-dimensional images and obtains the three-dimensional coordinates of the upper and lower edges of the lesion by specifying the Top and Bottom points in three-dimensional volume data.

[0010] The adaptive segmentation module is used to construct multi-scale ROIs at interaction points, expanding them step by step according to physical size or pixel size. After removing the lowest percentile gray level in each candidate ROI, the mean and standard deviation are calculated, and an adaptive threshold is obtained. Combined with filtering, morphological opening and closing operations, and aspect ratio and area filtering of connected components, the target connected regions containing interaction points are extracted. Abnormal results that are in contact with the ROI boundary are removed by edge detection. Iterative optimization is performed between different ROI scales to obtain the segmentation mask of the current layer.

[0011] The 3D interpolation module is used to perform linear interpolation along the layer direction based on the 3D coordinates of the Top and Bottom points in the 3D volume data, generate interpolation seed points for each intermediate layer within the Top–Bottom range, and call the adaptive segmentation module layer by layer to obtain a volume-coherent 3D segmentation mask.

[0012] The visualization export module is used to overlay the segmentation contours, seed points, and Top / Bottom markers onto the original image after window width and window level adjustments, and to export the 2D mask and overlay image as PNG or JPG files, and to export the 3D segmentation mask along with the voxel spacing information as an NIfTI file.

[0013] Preferably, the parameter persistence module uses QSettings to store user-defined parameters, enabling cross-session loading and reuse of segmented parameters, and providing a function to restore built-in default values.

[0014] Preferably, the interactive point selection module implements mouse settings based on the PyQt graphics event mechanism. In 2D mode, users can set seed points by clicking with the mouse and adjust the window width, window position, and zoom by dragging. In 3D mode, it supports scrolling or using a slider to browse layers, and users can enter the top and bottom edge annotation modes by using the "Top / Bottom" buttons respectively. Clicking on the corresponding layer records the three-dimensional coordinates. The interface automatically enables the corresponding controls according to the current data type and displays the current mode and coordinates in the status bar.

[0015] Preferably, the adaptive segmentation module establishes a multi-scale ROI at the interaction point and determines a threshold based on local statistics to automatically obtain a two-dimensional segmentation mask as follows:

[0016] A1. Using the seed point as the center, generate candidate square ROIs from smallest to largest according to physical size or pixel size;

[0017] A2. Sort the gray levels in each ROI from smallest to largest, remove the pixels with the lowest gray level according to a preset ratio, calculate the mean and standard deviation of the remaining pixels, and obtain the adaptive threshold by thr=mean+K_STD×std, where K_STD is a constant coefficient used to control the sensitivity of the threshold to outliers.

[0018] A3. Perform threshold segmentation and median filtering on the ROI, and open / close operations on elliptical structuring elements to obtain a smooth binary mask.

[0019] A4. Perform connected component analysis, and remove abnormal connected components according to the ratio of major axis to minor axis and area, where the area is the physical area or pixel area.

[0020] A5. In the remaining connected regions, prioritize the connected regions containing the seed point as candidate lesions. If none are found, select the one with the largest area as the candidate lesion.

[0021] A6. Check if the candidate connected component touches the ROI boundary. If it touches the boundary, the scale is considered unsuitable. Discard the result of that scale and try a larger ROI. Once a connected component that "contains the seed point and does not touch the boundary" is obtained, stop the iteration and map it back to the original image coordinates as the final mask. If all scales fail, use the result of the last ROI as the fallback output.

[0022] Preferably, the three-dimensional interpolation module performs linear interpolation based on the positions of the top and bottom points, generates interpolation seed points for each layer within the Top–Bottom range, and calls the adaptive ROI segmentation module layer by layer to obtain the segmentation results of continuous layers.

[0023] Preferably, the visualization export module uses the grayscale image as the background under the current window width and window level, and displays the segmentation contour, seed points, and Top or Bottom markers in different colors, with additional left and right labels.

[0024] This invention also discloses a medical image segmentation method based on interactive point selection and adaptive ROI, specifically including the following steps:

[0025] Step 1: Read the input image file and automatically identify its type;

[0026] Step 2: Receive the seed point, vertex, or bottom point set by the user;

[0027] Step 3: Expand the ROI region incrementally, by pixel or centimeter.

[0028] Step 4: After removing the lowest percentile gray level, calculate the mean and standard deviation to obtain the dynamic threshold;

[0029] Step 5: Perform morphological opening and closing operations, aspect ratio and area threshold filtering;

[0030] Step 6: Extract connected regions containing interaction points;

[0031] Step 7: Output the mask result.

[0032] Preferably, in step three, the ROI expansion scale is automatically converted to centimeters based on the pixel spacing to ensure that the physical size of different images is consistent.

[0033] Preferably, the segmentation results are automatically excluded from abnormal masks and iteratively optimized through edge detection and statistical feedback. Beneficial effects

[0034] This invention provides a medical image segmentation system and method based on interactive point selection and adaptive ROI adjustment. Compared with existing technologies, it has the following advantages:

[0035] 1. This medical image segmentation system and method based on interactive point selection and adaptive ROI control automatically completes ROI extraction and threshold calculation through a single user click. It uses a statistical model to dynamically determine the grayscale threshold and perform morphological optimization to achieve high-precision segmentation of two-dimensional and three-dimensional images. The system supports multi-format medical image input (DICOM, NIfTI, PNG, JPG, etc.) and uses a PyQt interface to achieve interactive segmentation, parameter persistence, and result visualization export. Compared with traditional thresholding methods and deep learning models, this invention has the advantages of strong adaptability, high robustness, and no need for training samples. It is suitable for scientific research and clinical auxiliary diagnosis of medical images such as MRI and CT.

[0036] 2. This medical image segmentation system and method based on interactive point selection and adaptive ROI control displays segmentation contours, seed points, and Top or Bottom markers in different colors, with added left and right labels. The 2D results can be exported as mask PNGs and overlaid PNG or JPG images, while the 3D results can be exported as NIfTI mask files with voxel spacing information for subsequent 3D display and quantitative analysis of volume, major axis, etc. The system can work uniformly in 2D and 3D modes, and the parameters can be persistently stored and automatically loaded. It has the advantages of simple operation, strong robustness, and cross-modal compatibility. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the medical image segmentation system of the present invention.

[0038] Figure 2 This is a system module structure diagram and an adaptive ROI threshold segmentation flowchart of the present invention.

[0039] Figure 3 This is a diagram illustrating the two-dimensional point selection and segmentation effect of the present invention.

[0040] Figure 4 This is a schematic diagram of the three-dimensional Top-Bottom first-person perspective effect of the present invention;

[0041] Figure 5 This is a schematic diagram of the three-dimensional Top-Bottom second-view effect of the present invention;

[0042] Figure 6 This is a schematic diagram of the corresponding software interface layout of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figures 1-6 This invention provides three technical solutions:

[0045] Example 1: A medical image segmentation system based on interactive point selection and adaptive ROI control, comprising:

[0046] The data input module is used to read medical images in various image formats, automatically parse pixel spacing, layer spacing, and window width and window level information, and normalize grayscale by default.

[0047] The parameter persistence module uses QSettings to save the adaptive threshold coefficient K_STD, the lowest percentile elimination ratio, the ROI start, the maximum size and step size, the connected component area and the slenderness ratio threshold, the drawing style and the export path segmentation and display parameters across sessions, and automatically loads them when the segmentation system starts.

[0048] The interactive point selection module, based on the PyQt graphics event mechanism, obtains seed points in two-dimensional images and obtains the three-dimensional coordinates of the upper and lower edges of the lesion by specifying the Top and Bottom points in three-dimensional volume data.

[0049] The adaptive segmentation module is used to construct multi-scale ROIs at interaction points, expanding them step by step according to physical size or pixel size. After removing the lowest percentile gray level in each candidate ROI, the mean and standard deviation are calculated, and an adaptive threshold is obtained. Combined with filtering, morphological opening and closing operations, and aspect ratio and area filtering of connected components, the target connected regions containing interaction points are extracted. Abnormal results that are in contact with the ROI boundary are removed by edge detection. Iterative optimization is performed between different ROI scales to obtain the segmentation mask of the current layer.

[0050] The 3D interpolation module is used to perform linear interpolation along the layer direction based on the 3D coordinates of the Top and Bottom points in the 3D volume data. It generates interpolation seed points for each intermediate layer within the Top–Bottom range and calls the adaptive segmentation module layer by layer to obtain a volume-coherent 3D segmentation mask.

[0051] The visualization export module is used to overlay the segmentation contours, seed points, and Top / Bottom markers onto the original image after adjusting the window width and window level, and to export the 2D mask and overlay image as PNG or JPG files, and to export the 3D segmentation mask along with the voxel spacing information as an NIfTI file.

[0052] The system automatically extracts ROI and calculates thresholds by a single user click, dynamically determines grayscale thresholds using a statistical model, and performs morphological optimization to achieve high-precision segmentation of two-dimensional and three-dimensional images. It supports multi-format medical image input (DICOM, NIfTI, PNG, JPG, etc.) and uses a PyQt interface for interactive segmentation, parameter persistence, and result visualization export. Compared with traditional thresholding methods and deep learning models, this invention has advantages such as strong adaptability, high robustness, and no need for training samples, making it suitable for research and clinical auxiliary diagnosis of medical images such as MRI and CT.

[0053] In this embodiment of the invention, the parameter persistence module uses QSettings to store user-defined parameters, enabling cross-session loading and reuse of segmented parameters, and providing a function to restore built-in default values.

[0054] In this embodiment of the invention, the interactive point selection module implements mouse settings based on the PyQt graphics event mechanism. In 2D mode, users can set seed points by clicking with the mouse and adjust the window width, window position, and zoom by dragging. In 3D mode, users can browse layers using the scroll wheel or slider and enter the top or bottom edge annotation mode using the "Top / Bottom" button respectively. Clicking on the corresponding layer records the three-dimensional coordinates. The interface automatically enables the corresponding control according to the current data type and displays the current mode and coordinates in the status bar.

[0055] In this embodiment of the invention, the adaptive segmentation module establishes a multi-scale ROI at the interaction point and determines the threshold based on local statistics to automatically obtain a two-dimensional segmentation mask.

[0056] A1. Using the seed point as the center, generate candidate square ROIs from smallest to largest according to physical size or pixel size;

[0057] A2. Sort the gray levels in each ROI from smallest to largest, remove the pixels with the lowest gray level according to a preset ratio, calculate the mean and standard deviation of the remaining pixels, and obtain the adaptive threshold by thr=mean+K_STD×std, where K_STD is a constant coefficient used to control the sensitivity of the threshold to outliers.

[0058] A3. Perform threshold segmentation and median filtering on the ROI, and open / close operations on elliptical structuring elements to obtain a smooth binary mask.

[0059] A4. Perform connected component analysis, and remove abnormal connected components according to the ratio of major axis to minor axis and area, where the area is the physical area or pixel area.

[0060] A5. In the remaining connected regions, prioritize the connected regions containing the seed point as candidate lesions. If none are found, select the one with the largest area as the candidate lesion.

[0061] A6. Check if the candidate connected component touches the ROI boundary. If it touches the boundary, the scale is considered unsuitable. Discard the result of that scale and try a larger ROI. Once a connected component that "contains the seed point and does not touch the boundary" is obtained, stop the iteration and map it back to the original image coordinates as the final mask. If all scales fail, use the result of the last ROI as the fallback output.

[0062] In this embodiment of the invention, the three-dimensional interpolation module performs linear interpolation based on the positions of the top and bottom points, generates interpolation seed points for each layer within the Top–Bottom range, and calls the adaptive ROI segmentation module layer by layer to obtain the segmentation results of continuous layers.

[0063] In this embodiment of the invention, the visualization export module uses the grayscale image as the background under the current window width and window level, and displays the segmentation contour, seed points, and Top or Bottom markers in different colors, with additional left and right labels.

[0064] By overlaying segmented contours, seed points, and Top or Bottom markers in different colors and adding left and right labels, the 2D results can be exported as mask PNGs and overlaid PNG or JPG images, while the 3D results can be exported as NIfTI mask files with voxel spacing information for subsequent 3D display and quantitative analysis of volume, major axis, etc. The system can work uniformly in 2D and 3D modes, and parameters can be persistently stored and automatically loaded. It has the advantages of simple operation, strong robustness, and cross-modal compatibility.

[0065] The 3D interpolation module uses Top / Bottom annotations in 3D data to generate volumetrically coherent segmentation results:

[0066] B1. Obtain the three-dimensional coordinates (zt, xt, yt) and (zb, xb, yb) of two points from the Top and Bottom annotations, and convert them to cm / px according to the pixel spacing and layer spacing in DICOM / NIfTI.

[0067] B2. Linearly interpolate the x and y coordinates in the z direction layer by layer to generate an interpolation seed point for each layer between Top and Bottom.

[0068] B3. The adaptive segmentation module is called layer by layer, with the interpolation seed point and the corresponding cm / px as input, to perform local segmentation on each layer and obtain the hierarchical mask;

[0069] B4. Stack the masks according to their layer numbers to form a three-dimensional mask volume, and construct an affine matrix by combining the voxel spacing. This ensures that the mask is physically aligned with the original volume data after being exported from NIfTI, thus achieving coherent segmentation of the lesion volume.

[0070] Example 2 differs from Example 1 in that: the present invention also discloses a medical image segmentation method based on interactive point selection and adaptive ROI, specifically including the following steps:

[0071] Step 1: Read the input image file and automatically identify its type;

[0072] Step 2: Receive the seed point, vertex, or bottom point set by the user;

[0073] Step 3: Expand the ROI region incrementally, by pixel or centimeter.

[0074] Step 4: After removing the lowest percentile gray level, calculate the mean and standard deviation to obtain the dynamic threshold;

[0075] Step 5: Perform morphological opening and closing operations, aspect ratio and area threshold filtering;

[0076] Step 6: Extract connected regions containing interaction points;

[0077] Step 7: Output the mask result.

[0078] In this embodiment of the invention, the ROI expansion scale in step three is automatically converted to centimeters based on the pixel spacing to ensure that the physical size of different images is consistent.

[0079] In this embodiment of the invention, the segmentation results are automatically eliminated by touch edge detection and statistical feedback, and then iteratively optimized.

[0080] The present invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a medical image segmentation method based on interactive point selection and adaptive ROI control.

[0081] Example 3 differs from Examples 1 and 2 in that this invention selects 148 MRI images and compares the performance metrics of traditional threshold segmentation (OTSU), k-means clustering, region growing, and the algorithm of this invention with a tolerance of 1 mm. The results are shown in Table 1.

[0082] Table 1. Comparison of Four Segmentation Methods (Edge-based Indicators) (mean±std, IQR)

[0083]

[0084] As can be seen from the results in the table above, the method of the present invention can achieve high performance in Dice, without requiring any training samples.

[0085] EdgeF1 achieved the highest mean and lowest variance in overall performance metrics, and also performed best in the HD95 distance error metric, proving that this method is superior to traditional thresholding methods, clustering algorithms and region growing methods in terms of segmentation accuracy and stability.

[0086] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A medical image segmentation system based on interactive point selection and adaptive ROI regulation, characterized in that: Comprise: ​ Data input module for reading a variety of image formats medical image, automatic analysis of pixel spacing, layer spacing and window width window level information, and in the default gray scale normalization processing; Parameter persistence module, based on QSettings cross-session save adaptive threshold coefficient K_STD, the lowest percentage of rejection ratio, ROI start, maximum size and step, connected domain area and elongation ratio threshold, drawing style and export path segmentation and display parameters, and automatically loaded when the segmentation system starts; Interactive point selection module, based on PyQt graphics event mechanism, in two-dimensional image to obtain seed point, in three-dimensional body data through the specified Top point and Bottom point to obtain the three-dimensional coordinates of the upper and lower edges of the lesion; Adaptive segmentation module, for constructing multi-scale ROI at the interactive point, expanding by physical size or pixel size step by step, calculating the mean and standard deviation after removing the lowest percentage of gray in each candidate ROI, and obtaining the adaptive threshold, combining filtering, morphological opening and closing operation and connected domain elongation ratio and area filtering, extracting the target connected region containing the interactive point, and removing the abnormal results touching the ROI boundary through edge detection, and iterating optimization between different ROI scales to obtain the segmentation mask of the current layer; Three-dimensional interpolation module, for linear interpolation in three-dimensional body data based on the three-dimensional coordinates of the Top point and Bottom point along the layer direction, generating interpolation seed points for each intermediate layer within the Top-Bottom range, and calling the adaptive segmentation module layer by layer to obtain the three-dimensional segmentation mask with continuous volume; Visual export module, for superimposed display of the segmentation contour and seed point, Top / Bottom label on the original image after window width window level adjustment, and export of two-dimensional mask and overlay as PNG or JPG file, and export of the three-dimensional segmentation mask together with the voxel spacing information as NIfTI file. 2.The medical image segmentation system based on interactive point selection and adaptive ROI regulation according to claim 1, wherein: The parameter persistence module uses QSettings to store user-defined parameters, realizes cross-session loading and reuse of segmentation parameters, and provides a function to restore built-in default values. 3.The medical image segmentation system based on interactive point selection and adaptive ROI regulation according to claim 1, wherein: The interactive point selection module realizes mouse setting operation based on PyQt graphics event mechanism. In 2D mode, the user sets the seed point by clicking the mouse, and adjusts the window width window level and zooms by dragging. In 3D mode, it supports scrolling or sliding to browse the layer, and enters the upper edge or lower edge labeling mode with "Top / Bottom" button respectively. Click to record three-dimensional coordinates on the corresponding layer. The interface automatically enables the corresponding controls according to the current data type, and prompts the current mode and coordinates in the status bar. 4.The medical image segmentation system based on interactive point selection and adaptive ROI regulation according to claim 1, wherein: The adaptive segmentation module establishes multi-scale ROI at the interactive point and determines the threshold based on local statistics. The operation of automatically obtaining two-dimensional segmentation mask is as follows: A1, generate candidate square ROIs from small to large in physical size or pixel size with the seed point as the center; A2, order the gray scale in each ROI from small to large, eliminate the lowest gray scale pixels of a preset proportion, calculate the mean and standard deviation of the remaining pixels, and obtain the adaptive threshold value according to thr = mean + K_STD * std, and K_STD is a constant coefficient for controlling the sensitivity of the threshold value to outliers; A3, perform threshold segmentation, median filtering and open-close operation of an elliptical structure element on the ROI to obtain a smooth binary mask; A4, perform connected component analysis, and eliminate abnormal connected components according to the length-to-short-axis ratio and area, and the area is a physical area or a pixel area; A5, select the connected component containing the seed point as the candidate lesion in the remaining connected components, and if not, select the connected component with the largest area as the candidate lesion; A6, detect whether the candidate connected component touches the ROI boundary, if it does, it is considered that the scale is not in conformity, the scale result is discarded and a larger ROI is tried, once a connected component "containing a seed point and not touching the boundary" is obtained, iteration is stopped, it is mapped back to the original image coordinates as the final mask, and if all scales fail, the result of the last ROI is used as the bottom output. 5.The medical image segmentation system based on interactive point selection and adaptive ROI regulation according to claim 1, wherein: The three-dimensional interpolation module performs linear interpolation according to the positions of the upper and lower points, generates interpolation seed points for each layer within the Top-Bottom range, and calls the adaptive ROI segmentation module layer by layer to obtain the segmentation results of the continuous layers. 6.The medical image segmentation system based on interactive point selection and adaptive ROI regulation according to claim 1, wherein: The visualization export module displays the segmentation contour, seed point and Top or Bottom label in different colors superimposed on the gray scale image as the background under the current window width and window position, and adds left and right identifiers.

7. The medical image segmentation method based on interactive point selection and adaptive ROI regulation, using the medical image segmentation system based on interactive point selection and adaptive ROI regulation according to any one of claims 1-6, characterized in that: Specifically, the following steps are included: Step one, read the input image file and automatically identify its type; Step two, receive the user-set seed point; Step three, expand the ROI area in pixel or centimeter units; Step four, calculate the mean and standard deviation after eliminating the lowest percentage gray scale in each candidate ROI, and obtain the adaptive threshold value; Step five, perform morphological open-close operation, elongation ratio and area threshold filtering; Step six, extract the connected region containing the interaction point; Step seven, output the mask result. 8.The medical image segmentation method based on interactive point selection and adaptive ROI control according to claim 7, characterized in that: The ROI expansion scale in step three is automatically converted into centimeter units according to the pixel spacing to ensure the consistency of the physical dimensions of different images. 9.The medical image segmentation method based on interactive point selection and adaptive ROI regulation according to claim 7, characterized in that: The segmentation result is automatically excluded from the abnormal mask through edge detection and statistical feedback and iteratively optimized.

Images