Ultrasonic image focus automatic identification method based on deep learning

Abstract

The invention provides an ultrasonic image focus automatic identification method based on deep learning, and the method comprises the steps: obtaining an ultrasonic probe pressure and an ultrasonic image of a body surface contact region of a patient, carrying out the tissue hierarchy analysis of the ultrasonic image, and recognizing the tissue density and the three-dimensional space coordinate information of the depth position of a target organ; driving sound wave signals of the ultrasonic probe through the target probe frequency, monitoring the contact pressure change of the probe and the body surface in real time, and judging whether the current contact state meets the pressure requirement of deep imaging or not according to the contact pressure change; identifying a deep focus boundary of the enhanced ultrasonic image, and performing multi-scale feature extraction and texture analysis on the ultrasonic image to obtain contour coordinates, tissue gray level distribution and internal uniformity of a focus area; meanwhile, the image noise level and the tissue contrast difference are evaluated to obtain the definition, the contrast ratio and the boundary sharpening degree of the current image.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an automatic identification method for lesions in ultrasound images based on deep learning. Background Technology

[0002] Ultrasound examination is an indispensable tool in medical diagnosis. Its non-invasive, real-time, and low-cost characteristics make it crucial for disease screening and diagnosis, especially in observing organ structure and hemodynamics. However, current ultrasound methods cannot adaptively adapt to differences in patient body size or tissue depth, leading to decreased image quality, particularly in obese patients or deep lesion examinations, where blurred boundaries and loss of detail are especially prominent. In ultrasound examinations, the attenuation of sound wave penetration is closely related to the diversity of tissue density. Differences in patient body size result in variations in the thickness and density of subcutaneous fat, muscle, and other tissue layers, causing uneven energy loss during sound wave propagation, often weakening the echo signal of deep lesions. More critically, existing equipment struggles to dynamically adjust probe parameters or signal processing methods based on real-time imaging feedback, making it difficult to clearly present details such as deep blood vessel distribution or tissue texture in patients with complex body shapes. For example, when examining the liver of an obese patient, ultrasound waves must penetrate a thick layer of fat, resulting in severe energy attenuation, blurred lesion boundaries, and even the inability to accurately distinguish between benign and malignant lesions. Because patients vary in body shape and tissue characteristics, fixed parameters cannot meet the signal enhancement requirements of deep imaging, while real-time adjustment of probe pressure or frequency requires complex signal feedback mechanisms. Therefore, optimizing the fit between sound wave penetration and imaging quality by acquiring patient body shape data, probe pressure signals, and ultrasound image sequences in a dynamically changing tissue environment has become a key issue in improving the accuracy of diagnosing deep lesions. Summary of the Invention

[0003] This invention provides a method for automatic identification of lesions in ultrasound images based on deep learning, mainly comprising: The process involves acquiring ultrasound probe pressure and ultrasound images at the contact area with the patient's body surface, performing tissue layer analysis on the ultrasound images to identify the three-dimensional spatial coordinates of tissue density and the depth location of the target organ, calculating the acoustic energy loss at different depths based on the three-dimensional spatial coordinates of the target organ's depth location and the ultrasound probe pressure, and adjusting the target probe frequency according to the tissue density. The ultrasound probe's acoustic signal is driven by the target probe frequency, and the contact pressure changes between the ultrasound probe and the body surface are monitored in real time. The contact pressure changes are used to determine whether the contact state meets the requirements for deep imaging. Edge detection is performed on the ultrasound images. The process involves measurement and signal enhancement processing to acquire enhanced ultrasound images; multi-scale feature extraction and texture analysis are performed on the enhanced ultrasound images to obtain the contour coordinates, tissue grayscale distribution, and internal homogeneity of the lesion region; the sharpness, contrast, and boundary sharpness of the enhanced ultrasound images are evaluated; the depth range of sound wave arrival is identified based on the sharpness, contrast, and boundary sharpness; the echo intensity data of each tissue layer is determined based on the contour coordinates, tissue grayscale distribution, and internal homogeneity of the lesion region; and the sound wave penetration is determined based on the depth range of sound wave arrival and the echo intensity data of each tissue layer, thereby acquiring ultrasound images that meet the requirements for deep imaging.

[0004] Furthermore, the acquisition of ultrasound probe pressure and ultrasound images in the contact area with the patient's body surface, and the analysis of tissue layers in the ultrasound images to identify the three-dimensional spatial coordinates of tissue density and the depth location of the target organ, includes: The process involves acquiring pressure distribution data of the area where the ultrasound probe contacts the patient's body surface, collecting pressure values ​​at different locations on the ultrasound probe using a piezoelectric sensor array, calculating the pressure gradient of the pressure distribution data, adjusting the ultrasound probe angle based on the pressure gradient, and acquiring an ultrasound image sequence under uniform contact. The ultrasound image sequence is then analyzed frame-by-frame to calculate the difference in grayscale values ​​between adjacent pixels, identify the boundaries between the subcutaneous fat layer, muscle layer, and organ parenchyma, determine the tissue type based on the mean and standard deviation of grayscale values ​​for each layer, match the acoustic characteristic database, and obtain the density and thickness parameters of each tissue layer. Based on the density and thickness parameters of each tissue layer, the depth distance between the upper and lower boundaries of the target organ is measured using ultrasound echo time. Combined with the ultrasound probe scanning angle and the ultrasound beam propagation path, the three-dimensional spatial coordinate information of the target organ is calculated.

[0005] Furthermore, the step of calculating the sound wave energy loss at different depth levels based on the three-dimensional spatial coordinates of the target organ's depth location and the ultrasound probe pressure, and adjusting the target probe frequency according to the tissue density, includes: Based on the three-dimensional spatial coordinates of the target organ's depth location, the thickness data and acoustic impedance values ​​of each tissue layer are extracted. The contact pressure distribution is obtained through an ultrasonic probe pressure sensor, and the sound beam propagation angle is calculated. Combining the thickness and acoustic attenuation coefficient of each tissue layer, the energy attenuation value of the sound wave propagating in each tissue layer is calculated. The energy attenuation values ​​are accumulated to obtain the total energy loss of the sound wave reaching the target organ depth. Based on the total energy loss and the tissue density, the frequency penetration performance database is queried to determine the suitable frequency range. Based on the frequency range, the excitation signal frequency of the ultrasonic probe transducer is adjusted, a test pulse is emitted, and the signal-to-noise ratio of the echo signal is monitored to determine the target probe frequency.

[0006] Furthermore, the step of driving the ultrasonic probe's acoustic signal at the target probe frequency, monitoring the change in contact pressure between the ultrasonic probe and the body surface in real time, and determining whether the contact state meets the requirements for deep imaging based on the change in contact pressure includes: A pulse excitation signal is generated by the target probe frequency to drive the ultrasonic probe transducer array to emit sound waves. The contact pressure value between the ultrasonic probe and the body surface is collected in real time, and the pressure gradient and pressure distribution uniformity are calculated to generate a pressure distribution map. Based on the pressure distribution map, areas with pressure values ​​below the deep imaging threshold are identified, and the tilt degree of the ultrasonic probe relative to the body surface and the pressure uniformity index are calculated. The direction of the ultrasonic probe is adjusted according to the location of the low-pressure area, and the pressure distribution data is collected again to determine whether the contact state meets the requirements for deep imaging.

[0007] Furthermore, the step of performing edge detection and signal enhancement processing on the ultrasound image to obtain an enhanced ultrasound image includes: Edge detection is performed on the ultrasound image, gray-level gradient values ​​of adjacent pixels are calculated, the clarity of deep tissue contours is identified, the penetration pressure range is determined according to the tissue density distribution, the pressure intensity and angle offset of the ultrasound probe are adjusted, and an adjusted ultrasound image sequence is obtained. Gain compensation processing is performed on the ultrasound image sequence, an increasing gain coefficient is applied, a histogram equalization algorithm is used to enhance the contrast of deep tissues, the gray-level distribution range is adjusted, noise is removed, and an enhanced ultrasound image is obtained.

[0008] Furthermore, the enhanced ultrasound image undergoes multi-scale feature extraction and texture analysis to obtain the contour coordinates, tissue grayscale distribution, and internal uniformity of the lesion region, including: The enhanced ultrasound image is decomposed into Gaussian pyramids to construct image sequences at different resolution levels. The Laplacian operator is applied to detect edge response intensity and determine the main contour location of the lesion. Based on the main contour location, texture features are extracted using a gray-level co-occurrence matrix, and contrast, correlation, energy, and entropy parameters are calculated to obtain the texture feature vector of the lesion region. According to the texture feature vector and the main contour location, a region growing algorithm is applied to expand the lesion region to obtain a contour coordinate sequence. The mean, standard deviation, and kurtosis of the pixels within the contour coordinate sequence are statistically analyzed to determine the tissue gray-level distribution and internal uniformity.

[0009] Furthermore, the evaluation of the sharpness, contrast, and edge sharpness of the enhanced ultrasound image includes: The enhanced ultrasound image is divided into local regions, and the standard deviation and mean of the pixel grayscale value of each region are calculated to determine the local noise level. The noise level distribution of the entire image is statistically analyzed. The average grayscale value of the target tissue and background regions is extracted, the contrast value is calculated, the grayscale difference at the tissue boundary is measured, the gradient amplitude is obtained, the signal-to-noise ratio and boundary sharpness metric are calculated, and the sharpness, contrast and boundary sharpness of the enhanced ultrasound image are determined.

[0010] Furthermore, the step of identifying the sound wave depth range based on the clarity, contrast, and boundary sharpness, determining the echo intensity data of each tissue layer based on the contour coordinates, tissue grayscale distribution, and internal uniformity of the lesion area, determining the sound wave penetration based on the sound wave depth range and the echo intensity data of each tissue layer, and obtaining an ultrasound image that meets the requirements for deep imaging includes: The sound wave propagation depth is estimated based on the sharpness and edge sharpness, and the maximum depth reached by the sound wave is determined by combining the contrast attenuation curve. Based on the contour coordinates of the lesion area and the grayscale distribution of the tissue, the pixel grayscale values ​​of each depth layer are statistically analyzed to calculate the echo intensity dataset. Based on the echo intensity dataset, the echo intensity attenuation rate and sound wave penetration coefficient of adjacent layers are calculated, and ultrasound images that meet the deep imaging standards are screened. The image with the highest comprehensive score in terms of sharpness, contrast, and edge sharpness is selected.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an automatic lesion identification method for ultrasound images based on deep learning. By real-time monitoring of the contact pressure between the probe and the body surface, combined with tissue layer analysis and multi-scale feature extraction, the method dynamically adjusts the probe frequency, pressure intensity, and angular offset to optimize acoustic penetration and image clarity. First, the invention obtains the depth location and tissue density of the target organ through ultrasound image analysis, calculates acoustic energy loss, and adjusts the probe frequency accordingly to improve deep imaging quality. When blurred deep tissue contours or weak echo signals are detected, the pressure parameters are further adjusted and signal enhancement processing is performed to reduce echo attenuation. Through multi-scale feature extraction and texture analysis, the lesion boundary and tissue grayscale distribution are accurately identified, and image clarity, contrast, and boundary sharpness are evaluated. Finally, the acoustic penetration is determined, generating high-quality deep tissue ultrasound images suitable for patients with complex body shapes. This invention significantly improves the clarity of deep tissue imaging and the accuracy of lesion identification, optimizing the ultrasound diagnostic effect. Attached Figure Description

[0012] Figure 1 This is a flowchart of an automatic lesion identification method for ultrasound images based on deep learning, according to the present invention.

[0013] Figure 2 This is a schematic diagram of an automatic lesion identification method for ultrasound images based on deep learning according to the present invention. Detailed Implementation

[0014] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0015] like Figure 1-2 This embodiment of an automatic lesion identification method based on deep learning in ultrasound images may specifically include: Step S101: Obtain the ultrasound probe pressure and ultrasound image of the contact area on the patient's body surface, perform tissue layer analysis on the ultrasound image, and identify the three-dimensional spatial coordinate information of tissue density and target organ depth location.

[0016] Pressure distribution data within the contact area between the ultrasound probe and the patient's body surface is acquired. A piezoelectric sensor array collects pressure values ​​at different locations on the probe. The pressure gradient in the contact area is calculated using the pressure distribution data. The uniformity of probe contact is determined based on the pressure gradient. When the pressure gradient exceeds a preset threshold, the probe angle is adjusted to obtain an ultrasound image sequence under uniform contact conditions. Frame-by-frame analysis of the ultrasound image sequence is performed. The boundaries between the subcutaneous fat layer, muscle layer, and organ parenchyma are identified by calculating the difference in grayscale values ​​between adjacent pixels. The tissue type is determined based on the mean and standard deviation of grayscale values ​​for each layer. A preset acoustic characteristic database is used to match the sound velocity and attenuation coefficient of each tissue layer, obtaining the density and thickness parameters of each tissue layer. Based on the density and thickness parameters of each tissue layer, the depth distance between the upper and lower boundaries of the target organ is measured using ultrasound echo time. Combined with the probe scanning angle and the ultrasound beam propagation path, the lateral offset and longitudinal depth of the target organ within the patient's body are calculated using the triangulation principle, obtaining three-dimensional spatial position information including X-axis, Y-axis, and Z-axis coordinates.

[0017] Specifically, in one embodiment, a piezoelectric sensor array is arranged on the surface of the ultrasonic probe, forming a matrix of multiple sensing units. Each sensing unit independently measures the local pressure value, and the pressure signal is converted into a digital signal by an analog-to-digital converter. The pressure gradient is obtained by dividing the pressure difference between adjacent sensing units by the unit spacing. When the gradient value exceeds a preset threshold, it indicates that the probe is not uniformly attached to the body surface. At this time, a robotic arm is used to fine-tune the probe posture until the pressure values ​​of each sensing unit tend to be consistent.

[0018] Specifically, tissue layer identification in ultrasound images relies on the differences in acoustic characteristics of different tissues. The subcutaneous fat layer exhibits low echogenicity in ultrasound images, with grayscale values ​​typically between 30 and 80; the muscle layer shows moderate echogenicity, with grayscale values ​​ranging from 80 to 150; and organ parenchyma displays specific echo patterns depending on the organ type. By setting a grayscale threshold range and scanning image pixels line by line, when the grayscale difference between adjacent pixels exceeds the set threshold, it is marked as a tissue boundary point. Successive boundary points form tissue layer boundaries, allowing for the measurement of the thickness of each layer.

[0019] For example, an acoustic properties database pre-stores reference values ​​for the sound velocity and attenuation coefficient of major human tissues. The sound velocity in adipose tissue is approximately 1450 m / s, in muscle tissue approximately 1580 m / s, and in liver parenchyma approximately 1570 m / s. By matching the measured echo intensity with the database reference values, the actual density values ​​of each tissue layer are determined.

[0020] In one possible implementation, three-dimensional spatial positioning is achieved by measuring the time interval between the transmission and reception of the ultrasound pulse. The depth distance is obtained by multiplying the echo time at the upper boundary of the target organ by half the sound velocity of the corresponding tissue. The probe scanning angle is acquired in real time using a built-in gyroscope, and the lateral offset of the target organ is calculated using trigonometric functions in conjunction with the fan-shaped scanning range of the ultrasound beam. The Z-axis depth is directly determined by the echo time, while the X-axis and Y-axis coordinates are calculated jointly by the probe position sensor and the scanning angle, forming complete three-dimensional spatial coordinate information.

[0021] Step S102: Based on the three-dimensional spatial coordinates of the target organ's depth location and the ultrasound probe pressure, the sound wave energy loss at different depth levels is obtained. If the sound wave energy loss exceeds a preset loss threshold, the target probe frequency is adjusted according to the tissue density.

[0022] Based on the three-dimensional spatial coordinates of the target organ's depth location, thickness data and acoustic impedance values ​​of each tissue layer are extracted. The contact pressure distribution between the ultrasound probe and the body surface is obtained using a pressure sensor. The actual propagation angle of the sound beam is calculated based on the pressure distribution. Combining the tissue thickness and acoustic attenuation coefficient of each layer, the energy attenuation value of the sound wave propagating through the subcutaneous fat layer, muscle layer, and organ parenchyma layer is calculated. The energy attenuation values ​​are accumulated layer by layer to obtain the total energy loss of the sound wave reaching the target organ depth. If the total energy loss exceeds a preset loss threshold, the spectral characteristics of the current probe's operating frequency are analyzed. Based on the differences in penetration ability of different frequencies in tissues of varying densities, a frequency range suitable for the current tissue structure is determined by querying a preset frequency penetration performance database. Based on this frequency range, the excitation signal frequency of the probe transducer is adjusted. Within the frequency range, the frequency value is adjusted incrementally with a preset step size. A test pulse is emitted, and the signal-to-noise ratio (SNR) of the echo signal is monitored. When the SNR reaches a preset standard and is higher than the previous frequency point, the current frequency value is recorded. Adjustment continues until the SNR no longer increases, and this frequency is determined as the target probe frequency. The ultrasonic beam is re-emitted using the target probe frequency to collect echo intensity data at different depths. The ratio of the echo intensity of each layer after frequency modulation to that before frequency modulation is calculated. The frequency adjustment effect is judged based on the ratio. If the increase in echo intensity of deep tissue is lower than the preset increase threshold, fine-tuning is performed based on the target probe frequency to obtain a working frequency that is suitable for the current tissue characteristics of the patient.

[0023] Specifically, in one implementation, the calculation of sound wave energy loss requires consideration of multiple factors. Three-dimensional spatial coordinate information contains the precise location of the target organ within the patient's body, where the Z-axis coordinate represents vertical depth, and the X and Y-axis coordinates represent lateral and longitudinal offsets. By extracting this coordinate data, the actual propagation path of the sound wave from the probe surface to the target organ is determined. A pressure sensor array collects the contact pressure distribution between the probe and the body surface in real time. The higher the pressure value, the better the coupling between the probe and the skin, and the smaller the energy loss when the sound wave is incident. When the pressure distribution is uneven, air gaps are generated in some areas, leading to enhanced sound wave reflection and a reduction in the actual sound wave energy entering the tissue.

[0024] Specifically, the acoustic attenuation coefficient is an important parameter measuring the rate of energy loss of sound waves propagating in different tissues. The acoustic attenuation coefficient for adipose tissue is approximately 0.48 dB / cm / MHz, for muscle tissue it is approximately 1.09 dB / cm / MHz, and for liver tissue it is approximately 0.5 dB / cm / MHz. As sound waves penetrate each tissue layer, the energy attenuates exponentially. The attenuation amount for that layer is obtained by multiplying the thickness of each layer by its corresponding attenuation coefficient and then by the current probe operating frequency. In obese patients with a thick subcutaneous fat layer, the remaining energy after sound waves penetrate the fat layer is significantly reduced, leading to decreased imaging quality of deep organs. The total energy loss value is obtained by accumulating the attenuation amounts of each tissue layer. When the total loss exceeds 20 dB, the echo signal strength is insufficient to form a clear image, triggering the frequency adjustment mechanism.

[0025] It should be noted that the frequency penetration performance database pre-stores data on the penetration characteristics of ultrasound waves at different frequencies in various tissues. Low-frequency ultrasound waves have good penetration capabilities but lower resolution, while high-frequency ultrasound waves have high resolution but limited penetration depth. Ultrasound waves at 2.5 MHz can effectively penetrate 15-20 cm in soft tissue, while those at 5 MHz can only penetrate 8-10 cm. The database records frequency points in the 1-10 MHz range at 0.5 MHz intervals, with each frequency point corresponding to the penetration depth and attenuation characteristics under different tissue densities. Based on the current patient's tissue density distribution, the corresponding frequency-penetration depth relationship curves are matched from the database to determine the frequency range that can reach the target depth while maintaining sufficient echo intensity.

[0026] Preferably, the frequency adjustment process employs a progressive search method. Within a defined frequency range, starting from the lowest frequency, the frequency is gradually increased in steps of 0.25 MHz. For each frequency adjustment, 10 test pulses are transmitted, echo signals are acquired, and the average signal-to-noise ratio (SNR) is calculated. The SNR is determined by the ratio of the peak power of the echo signal to the background noise power. When the SNR reaches 15 dB or higher, the signal quality is considered to meet the imaging requirements. The frequency is further increased while monitoring the SNR change; adjustment is stopped when the SNR begins to decrease or the increase is less than 1 dB. At this point, the frequency value achieves a balance between penetration depth and imaging resolution, and is determined as the target probe frequency.

[0027] In one possible implementation, the evaluation of the frequency modulation effect is achieved through comparative analysis. The same examination area is rescanned using the target probe frequency, and echo data is collected at each depth layer. The echo intensity after frequency modulation is compared layer by layer with that before frequency modulation, and the intensity improvement ratio is calculated. For superficial tissues, which are closer to the probe, frequency adjustment has a smaller impact, and the improvement ratio is usually within 10%. For deeper tissues, especially at the depth of the target organ, the echo intensity improvement should reach more than 30%. If the improvement ratio does not reach the preset threshold, it indicates that the current frequency has not yet reached its optimal state.

[0028] For example, the fine-tuning process is based on the principle of gradient optimization. Starting from the target probe frequency, the frequency is adjusted upwards and downwards by 0.1 MHz, and the changes in echo intensity in both directions are compared. The direction with the largest increase in intensity is selected for further fine-tuning until the intensity no longer increases. This fine adjustment ensures that the frequency parameters are precisely matched to the individual tissue characteristics of the patient. Differences in tissue structure among different patients significantly affect frequency selection. Lean patients have thin subcutaneous fat, allowing for the use of higher frequencies to obtain high-resolution images. Obese patients require lower frequencies to enhance penetration, sacrificing some resolution but ensuring visibility of deep lesions. Through dynamic frequency adjustment, an adaptive match between ultrasound examination parameters and patient body shape is achieved.

[0029] For example, when examining the pancreas deep in the abdomen, a clear image can be obtained using a frequency of 3.5 MHz for patients of standard build. However, for obese patients with abdominal fat thickness exceeding 5 cm, the frequency needs to be reduced to 2.5 MHz or even 2.0 MHz to allow the sound waves to effectively penetrate the fat layer and reach the pancreas. The determination of the working frequency can also consider the differences in the purpose of the examination. If the main focus is on observing the overall morphology of the organ, the frequency can be appropriately reduced to improve penetration; if it is necessary to observe internal fine structures or vascular distribution, the frequency should be increased as much as possible while ensuring basic penetration.

[0030] Step S103: Drive the ultrasonic probe's acoustic signal through the target probe frequency, and simultaneously monitor the change in contact pressure between the probe and the body surface in real time. Determine whether the current contact state meets the pressure requirements for deep imaging based on the change in contact pressure.

[0031] A pulse excitation signal is generated at the target probe frequency to drive the ultrasound probe transducer array to emit sound waves. Simultaneously, a pressure sensor array is activated to collect the contact pressure values ​​between each area of ​​the probe and the patient's body surface in real time. The pressure gradient is obtained by dividing the pressure difference between adjacent sensors by the sensor spacing. The pressure distribution uniformity is evaluated by the standard deviation of the pressure values, resulting in a pressure distribution map of the current contact state. Regions with pressure values ​​below a preset deep imaging pressure threshold are identified based on this pressure distribution map. The probe's fit is determined by comparing the actual pressure value of each region with the threshold. If multiple consecutive sensors have pressure values ​​below the threshold, the coordinate range of that region is marked, obtaining the specific location information of the probe's poor fit. Based on this specific location information and pressure gradient data, the tilt of the probe relative to the body surface is calculated by the asymmetry of the pressure distribution. The pressure uniformity index is determined based on the ratio of the maximum pressure value to the minimum pressure value. When this index exceeds the preset uniformity threshold, the probe adjustment direction is determined based on the location of the low-pressure region. The pressure distribution data is then re-collected after adjustment to determine whether the current contact state meets the pressure requirements for deep imaging.

[0032] Specifically, in one embodiment, the target probe frequency is converted into a square wave signal of a specific frequency by a digital signal processor, amplified by a power amplifier, and then drives the transducer array. The transducers convert the electrical signal into mechanical vibrations, generating an ultrasonic beam of the corresponding frequency. The pressure sensor array uses thin-film piezoresistive sensors, distributed in a 5×5 matrix on the probe surface. Each sensor independently measures the local contact pressure, with a sampling frequency of 100Hz, to monitor dynamic pressure changes in real time.

[0033] Specifically, the pressure gradient is calculated using the finite difference method. The pressure difference between adjacent sensors is divided by the distance between the sensor centers to obtain the rate of pressure change in that direction. The lateral and longitudinal pressure gradients are calculated separately, forming a two-dimensional gradient vector field. The uniformity of pressure distribution is evaluated by calculating the standard deviation of the pressure values ​​from all sensors; a smaller standard deviation indicates a more uniform pressure distribution. When the standard deviation exceeds 20% of the average pressure value, the pressure distribution is considered non-uniform.

[0034] It should be noted that the pressure threshold for deep imaging is preset according to different examination sites. The pressure threshold is typically set to 8-12 kPa for abdominal examinations, 5-8 kPa for thyroid examinations, and 10-15 kPa for cardiac examinations. These thresholds are derived from a large amount of clinical data to ensure that the sound waves can fully penetrate the surface tissue to reach the target depth. When the pressure value in a certain area is lower than 80% of the corresponding threshold, that area is marked as a poorly fitted area. If three or more sensors show insufficient pressure consecutively, it indicates a significant coupling problem in that area.

[0035] For example, the probe tilt is calculated using the centroid offset of the pressure distribution. Using the pressure value as a weight, the centroid coordinates of the pressure distribution are calculated; the distance of the centroid from the probe's geometric center reflects the tilt. When the offset exceeds 30% of the probe radius, the probe angle needs adjustment. The pressure uniformity index is defined as the ratio of the maximum pressure value to the minimum pressure value. Ideally, this ratio is close to 1; in practice, a ratio exceeding 2 requires readjustment.

[0036] In one possible implementation, the probe adjustment direction is determined based on the spatial distribution of the low-pressure area. If the low-pressure area is concentrated at the probe tip, the probe is tilted forward; if it is distributed on the side, the angle is adjusted to the corresponding side. The adjustment amount is proportional to the pressure difference, and each adjustment angle is controlled within 5 degrees to avoid over-adjustment that could lead to loss of contact.

[0037] Preferably, after adjustment, pressure data is re-acquired to verify whether the requirements for deep imaging are met. When the pressure values ​​of all sensors are higher than the threshold and the standard deviation of the pressure distribution is less than 15% of the average value, the pressure requirements for deep imaging are considered to be met. At this time, the probe forms good acoustic coupling with the body surface, ensuring that the sound waves can effectively propagate to the target depth.

[0038] Step S104: If the current ultrasound image shows blurred outlines of deep tissues or weak echo signals, adjust the pressure and angle offset of the ultrasound probe according to the tissue density, obtain the ultrasound image after the ultrasound probe is adjusted, and perform signal enhancement processing on the ultrasound image to reduce the echo attenuation of deep tissues and obtain an enhanced ultrasound image.

[0039] Edge detection processing is performed on the current ultrasound image. The clarity of deep tissue contours is identified by calculating the gray-level gradient values ​​of adjacent pixels. If the gradient value is lower than a preset clarity threshold or the echo signal amplitude is lower than a preset intensity threshold, the tissue density distribution data for that depth layer is extracted. The optimal penetration pressure range is determined based on the correspondence between density values ​​and sound wave propagation speed. Based on the optimal penetration pressure range, the difference between the current probe pressure and the target pressure is calculated. The probe pressure is adjusted by gradually increasing or decreasing the applied force. Simultaneously, the probe angle offset direction is determined based on the asymmetry of the tissue density distribution. The probe tilt angle is gradually adjusted in a manner that each adjustment does not exceed a preset angle threshold, obtaining an adjusted ultrasound image sequence. Gain compensation processing is performed on the ultrasound image sequence. Based on the attenuation degree of sound waves at different depths, an increasing gain coefficient is applied to the deep signal. A histogram equalization algorithm is used to enhance the contrast of deep tissues. The visibility of weak echo signals is improved by adjusting the image gray-level distribution range, resulting in a signal-enhanced ultrasound image. For the enhanced ultrasound image, image data acquired from different scanning angles are fused, and speckle noise is reduced by weighted averaging. Median filtering is used to remove noise while preserving edge features. The enhancement effect is verified by the increase ratio of deep tissue echo intensity, resulting in an enhanced ultrasound image that clearly displays deep tissue structures.

[0040] Specifically, in one implementation, the clarity of ultrasound image contours is determined based on the edge detection principle of gradient operators. The Sobel operator is used to perform convolution operations on the image, calculating the gradient components in the horizontal and vertical directions respectively. The square root of the sum of the squares of the gradients in both directions yields the gradient amplitude. Gradient amplitudes in deep tissue regions are generally lower, indicating blurred tissue boundaries. When the average gradient amplitude is below a preset threshold, the contour is considered unclear. Echo signal intensity is evaluated by calculating the mean grayscale value of pixels within the region of interest. The echo intensity value of normal liver parenchyma is typically in the range of 100-150 grayscale levels; a detection value below 80 is considered a weak signal.

[0041] Specifically, the extraction of tissue density distribution data involves the mapping relationship of acoustic characteristic parameters. Different tissue densities have different effects on the propagation speed and attenuation of sound waves. Adipose tissue density is approximately 0.92 g / cm³. 3 This corresponds to a sound speed of 1450 m / s; muscle tissue density is 1.04 g / cm³. 3 This corresponds to a sound speed of 1580 m / s; liver tissue density is 1.06 g / cm³. 3The corresponding sound velocity is 1570 m / s. By analyzing the spectral characteristics of the echo signal, tissue type is identified and density values ​​are extracted. The acoustic impedance of each tissue layer is calculated using the acoustic impedance formula Z = ρ × c, where ρ is the tissue density and c is the sound velocity. The greater the difference in acoustic impedance, the stronger the interface reflection, and the greater the required penetration pressure. Higher density tissues require a pressure range of 8-12 kPa, while lower density tissues only require 4-8 kPa to penetrate.

[0042] It should be noted that the probe pressure adjustment uses a closed-loop control method for precise adjustment. The pressure sensor provides real-time feedback of the current pressure value, which is compared with the target pressure to generate a deviation signal. When the deviation is positive, the applied pressure is gradually reduced; when the deviation is negative, the applied pressure is increased. Each adjustment is controlled within 10% of the current pressure to avoid sudden changes that could cause the probe to detach from the body surface. The angle adjustment is based on the spatial distribution characteristics of tissue density. When the density distribution is higher on the left and lower on the right, the probe tilts to the right; when it is higher in the front and lower in the back, the probe tilts backward. The tilt angle is proportional to the density difference; for every 0.1 g / cm³ increase in density difference... 3 Adjust the angle to increase by 2 degrees.

[0043] For example, the gain compensation processing is designed based on the attenuation law of sound waves in tissue. Sound wave intensity decreases exponentially with propagation distance, and the degree of attenuation is related to frequency and tissue characteristics. For a probe frequency of 3.5 MHz, the attenuation is approximately 0.7 dB per centimeter in soft tissue. At a tissue depth of 10 cm, the cumulative attenuation reaches 7 dB. The gain compensation curve is set in segments according to depth: a gain coefficient of 1.0 for depths of 0-3 cm, 1.5 for 3-6 cm, 2.0 for 6-10 cm, and 2.5 for depths above 10 cm. By applying corresponding gain coefficients to the echo signals at different depths, energy loss during propagation is compensated, thus restoring the echo intensity of deep tissues.

[0044] Preferably, the histogram equalization algorithm enhances contrast by redistributing the gray levels of an image. It involves statistically analyzing the gray-level histogram of the original image, calculating the cumulative distribution function, and remapping pixels concentrated in a certain gray-level range to the entire gray-level space. For images of deep tissues, the original gray-level values ​​are mostly concentrated in the range of 30-80; after equalization, these values ​​are expanded to the full range of 0-255, making tissue details more clearly visible.

[0045] In one possible implementation, multi-angle image fusion employs a weighted overlay method. The same region is scanned from three angles: -15 degrees, 0 degrees, and +15 degrees, resulting in three images. Weight coefficients are assigned to each image based on its signal-to-noise ratio (SNR), with higher SNR resulting in higher weights. Typical weight assignments are 0.3, 0.4, and 0.3, with slightly higher weights for the middle angles.

[0046] Understandably, speckle noise is an inherent characteristic of ultrasound images, caused by coherent scattering of sound waves. Weighted averaging reduces random noise by fusing multiple frames. A median filter is used with a 3×3 or 5×5 window, replacing the center pixel value with the median of the pixels within the window, removing impulse noise while preserving edge information. Furthermore, the enhancement effect is validated using quantitative metrics. The average grayscale value of the deep tissue region before and after enhancement is calculated; an improvement of 30% or more is considered effective. Contrast improvement is also assessed using the Weber contrast formula C = (I_target - I_background) / I_background, where I_target is the target tissue intensity and I_background is the background intensity. A contrast improvement of over 50% indicates a significant improvement in image quality.

[0047] For example, in the examination of deep liver lesions, the original image showed a lesion with a depth of 8 cm and a blurred boundary, with a grayscale value of only 60. After pressure and angle adjustments, combined with signal enhancement processing, the grayscale value of the lesion area was increased to 120, the boundary became clearly discernible, and the internal echo characteristics were obvious, achieving a clear display of deep tissue structures.

[0048] Step S105: Identify the boundaries of deep lesions in the enhanced ultrasound image, and perform multi-scale feature extraction and texture analysis on the ultrasound image to obtain the contour coordinates, tissue grayscale distribution, and internal uniformity of the lesion area.

[0049] Gaussian pyramid decomposition was performed on the enhanced ultrasound images to construct image sequences at different resolution levels. The Laplacian operator was applied to each level to detect edge response intensity. By comparing the edge intensity values ​​at the same location in adjacent levels, the level with the largest response value was selected as the feature scale for that location. The main contour location of the lesion was determined based on the optimal scale distribution across all locations. Based on the main contour location, texture features were extracted using a gray-level co-occurrence matrix. Contrast, correlation, energy, and entropy parameters were calculated in the horizontal, vertical, and diagonal directions. The contrast value was used to determine the degree of difference between the lesion and surrounding tissues, the correlation was used to assess the consistency of texture direction, the energy reflected the uniformity of gray-level distribution, and the entropy value measured the randomness of the texture, resulting in a texture feature vector for the lesion region. Based on the texture feature vector and contour location, the pixel with the highest gray-level value within the contour was selected as the seed point. A region growing algorithm was applied to expand from the seed point. By comparing the average gray-level difference and texture similarity between adjacent pixels and the grown region, growth stopped when the difference exceeded a preset threshold, obtaining a complete pixel set of the lesion region. Chaincode tracing was performed on the boundary of the pixel set to obtain the contour coordinate sequence. For the lesion region within the contour coordinate sequence, the gray values ​​of all pixels within the region are statistically analyzed, and the gray mean, standard deviation, and fourth-order central moment are calculated. The internal gray uniformity is evaluated by the ratio of the standard deviation to the mean. The kurtosis value is obtained by dividing the fourth-order central moment by the fourth power of the standard deviation. Thus, the contour coordinates, tissue gray distribution characteristics, and internal uniformity index of the lesion region are obtained.

[0050] Specifically, in one implementation, Gaussian pyramid decomposition constructs a multi-resolution image sequence through layer-by-layer downsampling. The original enhanced ultrasound image serves as the bottom layer of the pyramid. After smoothing with a Gaussian filter, it is sampled every other pixel to obtain a second layer image with half the size. This process is repeated to construct a 4-5 layer pyramid structure, with each layer decreasing in resolution to correspond to different observation scales. The Laplacian operator performs edge detection on each layer, identifying locations with drastic gray-level changes by calculating the second derivative of a pixel with its eight neighboring pixels. Lesion boundaries are most prominent at a specific scale, which matches the lesion size. Small lesions show a strong response at high-resolution levels, while large lesions are more prominent at low-resolution levels. The gray-level co-occurrence matrix (GLCM) is an important tool for describing image texture features, constructed by statistically analyzing the frequency of gray-level value pairs at specific directions and distances in the image. The pixel spacing is set to 1, and the statistical directions include 0 degrees, 45 degrees, 90 degrees, and 135 degrees. For each direction, a 256×256 co-occurrence matrix is ​​constructed, where the matrix element P(i,j) represents the number of times a pixel with gray value i and a pixel with gray value j occur adjacently in a specified direction. Contrast reflects the sharpness of the texture, and its calculation involves summing the square of the gray-level difference and the probability of occurrence. The contrast difference between lesion tissue and normal tissue typically exceeds 30%. Correlation measures the linear dependence of gray-level values, with a value range between -1 and 1; positive values ​​indicate positive correlation, and negative values ​​indicate negative correlation. Energy reflects the uniformity of gray-level distribution; uniform textures have higher energy values. Entropy measures the randomness and complexity of the texture; complex textures have higher entropy values, while simple textures have lower entropy values. These four parameters together constitute the texture feature vector, comprehensively describing the texture characteristics of the lesion region.

[0051] It should be noted that the selection of the seed point directly affects the region growth effect. By scanning all pixels within the contour, the point with the highest gray value is selected as the initial seed. This point is usually located in the central region of the lesion and has typical lesion characteristics. During the region growth process, the difference in the mean gray value between the pixel to be grown and the already grown region is calculated. If the difference is less than a threshold, it is included in the growth region. At the same time, texture similarity is compared, and Euclidean distance is used to measure the difference in texture feature vectors. The growth threshold is dynamically adjusted according to the tissue type, set to 15 gray levels for liver tissue and 20 gray levels for kidney tissue.

[0052] For example, chain code tracing uses an 8-directional encoding method to record the contour. Starting from any point on the contour, the next boundary point is searched in a clockwise direction, with the 8 directions represented by numbers 0-7. After completely traversing the contour, a chain code sequence is obtained, which accurately records the shape information of the contour. The contour length, area, and shape complexity can be calculated using the chain code.

[0053] Preferably, statistical analysis of the lesion area provides quantitative assessment indicators. The mean gray value reflects the average echo intensity of the tissue; the mean gray value for normal liver parenchyma is approximately 120, decreasing to below 90 in fatty liver and rising to above 140 in cirrhosis. The standard deviation measures the dispersion of gray values; the standard deviation is smaller for homogeneous tissues and larger for non-homogeneous tissues.

[0054] In one possible implementation, the calculation of the fourth central moment involves the fourth power of the difference between the grayscale value and the mean. First, the difference between the grayscale value of all pixels and the mean is calculated. The sum of these fourth-powered differences is then divided by the total number of pixels to obtain the fourth central moment. The kurtosis value is equal to the fourth central moment divided by the fourth power of the standard deviation minus 3. Positive kurtosis indicates a more concentrated distribution than a normal distribution, while negative kurtosis indicates a flatter distribution. Internal uniformity can be evaluated using multiple parameters. The ratio of the standard deviation to the mean is called the coefficient of variation; less than 0.3 indicates high uniformity, 0.3-0.5 indicates moderate uniformity, and greater than 0.5 indicates non-uniformity. A kurtosis value close to 0 indicates a grayscale distribution close to a normal distribution, while a value greater than 3 indicates the presence of abnormally high or low value clusters.

[0055] Understandably, these quantitative indicators provide an objective basis for judging the benign or malignant nature of lesions. Benign lesions are usually characterized by clear boundaries, homogeneous interior, and regular texture; malignant lesions often exhibit blurred boundaries, uneven interior, and disordered texture.

[0056] For example, in the detection of hepatic hemangiomas, the lesions are round or oval, with clear and smooth borders, uniform internal echoes, a grayscale standard deviation of less than 15, a kurtosis value close to 0, and a texture energy value higher than 0.8. In contrast, hepatocellular carcinoma exhibits irregular shapes, blurred borders, heterogeneous internal echoes, a standard deviation greater than 25, a kurtosis value far deviating from 0, and a texture entropy value exceeding 2.5. Through comprehensive analysis of these characteristic parameters, accurate identification and quantitative assessment of the lesions can be achieved.

[0057] Step S106: Simultaneously evaluate the image noise level and tissue contrast difference to obtain the sharpness, contrast and edge sharpness of the current image.

[0058] The ultrasound image is divided into multiple local regions. The standard deviation and mean of the pixel grayscale values ​​in each region are calculated. The local noise level is obtained by the ratio of the standard deviation to the mean. A sliding window is used to traverse the entire image, and the noise level distribution of all windows is statistically analyzed. The overall noise level of the image is determined based on the median of the distribution. Based on the overall noise level, the average grayscale values ​​of the target tissue and background regions are extracted. The absolute value of the difference between the two is calculated as the contrast value. Simultaneously, the grayscale difference between adjacent pixels at the tissue boundary is measured to obtain the gradient amplitude. The signal-to-noise ratio (SNR) is obtained by dividing the contrast value by the noise level. When the SNR exceeds a preset threshold, the image sharpness is deemed acceptable. The contrast level is determined based on the contrast value. For the gradient amplitude, a grayscale change curve is extracted along the normal direction of the tissue boundary. The maximum grayscale difference between adjacent points on the curve is calculated as the boundary sharpness metric. The degree of boundary sharpness is determined by the ratio of the sharpness metric to the average grayscale difference of adjacent regions. Combining the sharpness determination result, the contrast level, and the degree of boundary sharpness, a comprehensive evaluation result of the current image's sharpness, contrast, and boundary sharpness is obtained.

[0059] Specifically, in one implementation, the assessment of local noise levels is achieved using statistical methods. The ultrasound image is divided into 16×16 pixel local regions, and the standard deviation σ and mean μ of the pixel grayscale values ​​are calculated within each region. The noise level is defined as the coefficient of variation CV = σ / μ, which eliminates the influence of brightness differences, making the noise levels of different regions comparable. A sliding window moves across the image in 8-pixel steps, ensuring 50% overlap between adjacent windows and improving the spatial resolution of the noise assessment.

[0060] Specifically, the noise distribution is statistically analyzed using a histogram method. Noise level values ​​for all windows are collected, and a noise level histogram is constructed. The median serves as a robust estimate of the overall noise level, unaffected by individual outliers. A median less than 0.1 is considered low noise, 0.1-0.2 is considered moderate noise, and greater than 0.2 is considered high noise.

[0061] It's important to note that signal-to-noise ratio (SNR) calculation involves the accurate identification of both the target tissue and the background. The target tissue is typically selected as the lesion area or organ of interest, while the background is selected as the surrounding normal tissue. The difference in their average grayscale values ​​reflects the contrast between the tissues. SNR = |I_target - I_background| / σ_noise, where σ_noise is the overall noise level. An SNR greater than 3 is considered acceptable in terms of sharpness, 1.5-3 is acceptable, and less than 1.5 indicates insufficient sharpness. Contrast levels are determined based on the grayscale difference: a difference greater than 50 indicates high contrast, 20-50 indicates medium contrast, and less than 20 indicates low contrast.

[0062] For example, the assessment of boundary sharpness is achieved through grayscale profile analysis. A one-dimensional grayscale curve, 20 pixels in length, is extracted along the tissue boundary normal. The grayscale difference between adjacent pixels is calculated, with the maximum difference characterizing the steepness of the boundary. The ratio of the sharpness metric to the average grayscale difference on both sides of the boundary reflects the sharpness of the boundary; a ratio closer to 1 indicates a sharper boundary, while a ratio closer to 0 indicates a more blurred boundary.

[0063] Preferably, the comprehensive evaluation result is obtained using a weighted average method. Clarity has a weight of 0.4, contrast has a weight of 0.3, and edge sharpness has a weight of 0.3. The three indicators are normalized to the range of 0-1, and then weighted and summed to obtain the comprehensive quality score. A score greater than 0.7 indicates excellent image quality, 0.5-0.7 is acceptable, and a score less than 0.5 requires readjustment of imaging parameters. Each acquired ultrasound image is evaluated immediately. When the quality score falls below the threshold, the operator is automatically prompted to adjust the probe position or imaging parameters to ensure high-quality diagnostic images are obtained.

[0064] Step S107: Identify the depth range of sound waves based on the clarity, contrast, and edge sharpness of the current image; determine the echo intensity data of each tissue layer based on the contour coordinates of the lesion area, tissue grayscale distribution, and internal uniformity; determine the sound wave penetration based on the depth range of sound waves and the echo intensity data of each tissue layer; and determine a high-quality deep tissue ultrasound image suitable for patients with complex body types based on the sound wave penetration.

[0065] Based on the current image's sharpness, contrast, and edge sharpness, the sound wave propagation depth is estimated by multiplying the sharpness value by the sound velocity constant. The edge sharpness is used to judge the imaging quality of each depth layer. When the sharpness is below a threshold, that depth is marked as the sound wave attenuation critical point. Combined with the contrast attenuation curve, the maximum actual depth reached by the sound wave is determined, obtaining the effective sound wave propagation depth range data. Based on the depth range data and the contour coordinates of the lesion region, the grayscale values ​​of each pixel within the contour are extracted. The main tissue type is determined based on the peak position of the tissue grayscale distribution. The acoustic characteristic differences of the tissue are identified through internal uniformity indicators. Pixel grayscale values ​​at different depth layers are statistically analyzed layer by layer, and the average grayscale value and standard deviation of each layer are calculated to obtain the echo intensity dataset for each tissue layer. Using the echo intensity dataset, the echo intensity attenuation rate between adjacent layers is calculated. The degree of sound wave energy loss is assessed based on the total attenuation within the sound wave's depth range. The sound wave penetration coefficient is determined by the ratio of the echo intensity of deep tissue to that of shallow tissue. When the penetration coefficient exceeds a preset threshold, the sound wave penetration is deemed to meet the requirements for deep imaging. Based on the sound wave penetration and the probe pressure and frequency parameters during image acquisition, a mapping relationship between imaging parameters and penetration is established. Ultrasound images that meet the deep imaging standard are screened out by the penetration value. The screened images are comprehensively scored according to clarity, contrast and boundary sharpness. The image with the highest score is selected as a high-quality deep tissue ultrasound image suitable for patients with complex body types.

[0066] Specifically, in one implementation, the identification of sound wave propagation depth is based on a multi-parameter comprehensive evaluation method. The sharpness value is obtained by calculating the modulation transfer function of the image, which describes the contrast transfer characteristics at different spatial frequencies. When the sharpness value is 0.8, the corresponding theoretical propagation depth is approximately 10 cm; when the sharpness value decreases to 0.5, the propagation depth increases to 15 cm. This correspondence is based on the attenuation law of sound waves in tissues; the higher the frequency, the faster the attenuation, and the shallower the propagation depth. The contrast attenuation curve exhibits an exponential decreasing trend, and is fitted with the curve equation y=ae^(-bx), where a is the initial contrast, b is the attenuation coefficient, and x is the depth value. When the contrast drops to 10% of the initial value, it is considered to have reached the limit depth for effective imaging. Boundary sharpness is used as an auxiliary criterion; when the sharpness degree is below 0.3, it indicates that image details at that depth have been severely lost, and this is marked as the critical point of sound wave attenuation.

[0067] Specifically, the acquisition of echo intensity data involves precise spatial localization and grayscale statistics. The lesion contour coordinates are obtained using a boundary tracing algorithm, forming a closed polygonal region. Within this region, the image is divided into several layers according to depth, with each layer having a thickness of 5 mm. All pixels within each layer are traversed, their grayscale values ​​are extracted, and stored in an array. The peak values ​​of the grayscale distribution are determined through histogram analysis; a unimodal distribution indicates a single tissue type, while a multimodal distribution indicates the presence of mixed tissue. Internal homogeneity is quantified by calculating the coefficient of variation (CV) = σ / μ, where σ is the standard deviation and μ is the mean. A CV less than 0.2 indicates highly homogeneous tissue, 0.2-0.4 indicates moderate homogeneity, and greater than 0.4 indicates non-homogeneity. The average grayscale value of each layer reflects the echo intensity at that depth, and the standard deviation reflects the dispersion of the echo. These statistical parameters are arranged in depth order to form a complete echo intensity dataset.

[0068] It should be noted that the assessment of acoustic wave penetration requires comprehensive consideration of both energy attenuation and signal quality. The formula for calculating the echo intensity attenuation rate α between adjacent layers is α = (I_n - I_n+1) / I_n × 100%, where I_n is the average echo intensity of the nth layer, and I_n+1 is the average echo intensity of the (n+1)th layer. The attenuation rate of normal soft tissue is approximately 5-10% per centimeter. When the attenuation rate exceeds 20%, it indicates the encounter of high-attenuation tissues such as calcified or fibrotic areas. The total attenuation is obtained by summing the attenuation values ​​of each layer, reflecting the total energy loss of the acoustic wave from the shallow to the deep layers. The acoustic wave penetration coefficient is defined as the ratio of the echo intensity of the deep tissue to the echo intensity of the shallow tissue. The closer this coefficient is to 1, the better the penetration effect. In practical applications, a penetration coefficient greater than 0.3 is sufficient for diagnostic needs, greater than 0.5 is good penetration, and greater than 0.7 is excellent penetration.

[0069] For example, the mapping relationship between imaging parameters and penetration power was established using experimental data. Within the probe pressure range of 5-15 kPa, each 1 kPa increase resulted in an approximately 2 mm increase in penetration depth. Within the probe frequency range of 2-5 MHz, each 0.5 MHz decrease in frequency resulted in an approximately 2 cm increase in penetration depth, but a corresponding decrease in image resolution. The relationship between pressure P, frequency f, and penetration power T was expressed as T = k1 × P / f + k2, where k1 and k2 are empirical coefficients derived through fitting with a large amount of clinical data.

[0070] Preferably, the image selection employs a multi-level filtering mechanism. The first level of filtering is based on the penetration value, eliminating images with a penetration coefficient less than 0.3. The second level of filtering is based on image quality indicators, requiring a sharpness greater than 0.5, a contrast ratio greater than 30, and a boundary sharpness greater than 0.4.

[0071] In one possible implementation, the overall score uses a weighted scoring method. Sharpness score S1 = Sharpness value × 100 × 0.35, Contrast score S2 = Contrast value / Maximum contrast × 100 × 0.35, Edge sharpening score S3 = Sharpening level × 100 × 0.3. The total score S = S1 + S2 + S3, with a maximum score of 100.

[0072] Understandably, this method is particularly suitable for examining deep organs in obese patients. When a patient's BMI exceeds 30, conventional ultrasound parameters are insufficient to obtain satisfactory images. By dynamically adjusting imaging parameters and optimizing penetration based on real-time feedback, even the posterior margin of the liver at a depth of up to 15 cm can be clearly displayed. Furthermore, the scoring weights can be adjusted for different examination sites. Cardiac examinations prioritize temporal resolution, reducing the weight for boundary sharpening to 0.2 and increasing the weight for clarity to 0.45. Examinations of abdominal parenchymal organs prioritize spatial resolution, maintaining the original weight allocation.

[0073] For example, when examining patients with severe fatty liver, the liver echo is significantly enhanced, with severe attenuation in deeper areas. Using this method, the effective imaging depth was automatically identified as 8 cm. By adjusting the probe frequency from 3.5 MHz to 2.5 MHz and the pressure from 10 kPa to 13 kPa, the penetration depth was increased to 12 cm, successfully displaying the deep structures of the liver and the course of the hepatic veins. The overall score improved from 45 to 78 points, meeting the requirements for clinical diagnosis.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based automatic lesion identification method for ultrasound images, characterized by, The method comprises the following steps: acquiring the ultrasound probe pressure and ultrasound image of the contact area of the patient's body surface, performing tissue hierarchical analysis on the ultrasound image, identifying the three-dimensional spatial coordinate information of the tissue density and the target organ depth position; calculating the sound wave energy loss at different depth levels according to the three-dimensional spatial coordinate information of the target organ depth position and the ultrasound probe pressure, adjusting the target probe frequency according to the tissue density; driving the sound wave signal of the ultrasound probe through the target probe frequency, monitoring the contact pressure change of the ultrasound probe and the body surface in real time, and judging whether the contact state meets the deep imaging requirement according to the contact pressure change; performing edge detection and signal enhancement processing on the ultrasound image to obtain an enhanced ultrasound image; performing multi-scale feature extraction and texture analysis on the enhanced ultrasound image to obtain the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area; evaluating the definition, contrast and boundary sharpening degree of the enhanced ultrasound image, identifying the sound wave reaching depth range according to the definition, contrast and boundary sharpening degree, determining the echo intensity data of each layer of tissue according to the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area, and determining the sound wave penetration according to the sound wave reaching depth range and the echo intensity data of each layer of tissue to obtain an ultrasound image meeting the deep imaging requirement. 2.The deep learning-based ultrasound image lesion automatic identification method according to claim 1, characterized in that, The method comprises the following steps: acquiring the ultrasound probe pressure and ultrasound image of the contact area of the patient's body surface, performing tissue hierarchical analysis on the ultrasound image, identifying the three-dimensional spatial coordinate information of the tissue density and the target organ depth position; calculating the sound wave energy loss at different depth levels according to the three-dimensional spatial coordinate information of the target organ depth position and the ultrasound probe pressure, adjusting the target probe frequency according to the tissue density; driving the sound wave signal of the ultrasound probe through the target probe frequency, monitoring the contact pressure change of the ultrasound probe and the body surface in real time, and judging whether the contact state meets the deep imaging requirement according to the contact pressure change; performing edge detection and signal enhancement processing on the ultrasound image to obtain an enhanced ultrasound image; performing multi-scale feature extraction and texture analysis on the enhanced ultrasound image to obtain the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area; evaluating the definition, contrast and boundary sharpening degree of the enhanced ultrasound image, identifying the sound wave reaching depth range according to the definition, contrast and boundary sharpening degree, determining the echo intensity data of each layer of tissue according to the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area, and determining the sound wave penetration according to the sound wave reaching depth range and the echo intensity data of each layer of tissue to obtain an ultrasound image meeting the deep imaging requirement. 3.The deep learning-based ultrasound image lesion automatic identification method according to claim 1, characterized in that, The method comprises the following steps: acquiring the ultrasound probe pressure and ultrasound image of the contact area of the patient's body surface, performing tissue hierarchical analysis on the ultrasound image, identifying the three-dimensional spatial coordinate information of the tissue density and the target organ depth position; calculating the sound wave energy loss at different depth levels according to the three-dimensional spatial coordinate information of the target organ depth position and the ultrasound probe pressure, adjusting the target probe frequency according to the tissue density; driving the sound wave signal of the ultrasound probe through the target probe frequency, monitoring the contact pressure change of the ultrasound probe and the body surface in real time, and judging whether the contact state meets the deep imaging requirement according to the contact pressure change; performing edge detection and signal enhancement processing on the ultrasound image to obtain an enhanced ultrasound image; performing multi-scale feature extraction and texture analysis on the enhanced ultrasound image to obtain the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area; evaluating the definition, contrast and boundary sharpening degree of the enhanced ultrasound image, identifying the sound wave reaching depth range according to the definition, contrast and boundary sharpening degree, determining the echo intensity data of each layer of tissue according to the contour coordinates, tissue gray scale distribution and internal uniformity of the lesion area, and determining the sound wave penetration according to the sound wave reaching depth range and the echo intensity data of each layer of tissue to obtain an ultrasound image meeting the deep imaging requirement. According to the three-dimensional space coordinate information of the target organ depth position, thickness data and acoustic impedance values of each tissue layer are extracted, contact pressure distribution is obtained through an ultrasonic probe pressure sensor, sound beam propagation angle is calculated, and energy attenuation values of sound waves in each tissue layer during propagation are calculated in combination with thickness and acoustic attenuation coefficients of each tissue layer; the energy attenuation values are accumulated to obtain total energy loss of sound waves reaching the target organ depth, and a frequency penetration performance database is queried according to the total energy loss and the tissue density to determine an adaptive frequency range; based on the frequency range, an excitation signal frequency of an ultrasonic probe transducer is adjusted, a test pulse is emitted, an echo signal signal-to-noise ratio is monitored, and a target probe frequency is determined. 4.The deep learning-based ultrasound image lesion automatic identification method of claim 1, wherein, The sound wave signal of the ultrasonic probe driven by the target probe frequency is used to monitor the contact pressure change of the ultrasonic probe and the body surface in real time, and whether the contact state meets the deep imaging requirement is judged according to the contact pressure change, including: A pulse excitation signal is generated through the target probe frequency to drive the ultrasonic probe transducer array to emit sound waves, the contact pressure value of the ultrasonic probe and the body surface is collected in real time, the pressure gradient and the pressure distribution uniformity are calculated, and a pressure distribution map is generated; according to the pressure distribution map, the region with a pressure value lower than the deep imaging threshold is identified, the inclination degree and the pressure application uniformity index of the ultrasonic probe relative to the body surface are calculated, the direction of the ultrasonic probe is adjusted according to the position of the low-pressure region, the pressure distribution data is collected again, and whether the contact state meets the deep imaging requirement is judged. 5.The deep learning-based ultrasound image lesion automatic identification method according to claim 1, characterized in that, The ultrasonic image is subjected to edge detection and signal enhancement processing to obtain an enhanced ultrasonic image, including: The ultrasonic image is subjected to edge detection, the adjacent pixel gray scale gradient value is calculated, the deep tissue profile definition is identified, the penetration pressure range is determined according to the tissue density distribution, the pressure intensity and angle offset of the ultrasonic probe are adjusted, and an adjusted ultrasonic image sequence is obtained; the ultrasonic image sequence is subjected to gain compensation processing, an incremental gain coefficient is applied, a histogram equalization algorithm is used to enhance the contrast of deep tissue, the gray scale distribution range is adjusted, and noise is removed to obtain an enhanced ultrasonic image. 6.The deep learning-based ultrasound image lesion automatic identification method according to claim 1, characterized in that, The enhanced ultrasonic image is subjected to multi-scale feature extraction and texture analysis to obtain the profile coordinates, tissue gray scale distribution and internal uniformity of the lesion region, including: The enhanced ultrasonic image is subjected to Gaussian pyramid decomposition to construct an image sequence of different resolution levels, a Laplacian operator is applied to detect the edge response strength, and the main profile position of the lesion is determined; based on the main profile position, a gray level co-occurrence matrix is used to extract texture features, contrast, correlation, energy and entropy parameters are calculated, and a texture feature vector of the lesion region is obtained; according to the texture feature vector and the main profile position, a region growing algorithm is applied to expand the lesion region to obtain a profile coordinate sequence, the gray mean, standard deviation and kurtosis value of the pixels in the profile coordinate sequence are counted, and the tissue gray scale distribution and internal uniformity are determined. 7.The deep learning-based ultrasound image lesion automatic identification method according to claim 1, characterized in that, The definition, contrast and boundary sharpening degree of the enhanced ultrasonic image are evaluated, including: The local region is divided for the enhanced ultrasound image, the standard deviation and the mean value of the pixel gray value of each region are calculated, the local noise level is determined, and the noise level distribution of the whole image is counted; the average gray value of the target tissue and the background region is extracted, the contrast value is calculated, the gray difference at the tissue boundary is measured, the gradient amplitude is obtained, the signal-to-noise ratio and the boundary sharpening metric value are calculated, and the definition, contrast and boundary sharpening degree of the enhanced ultrasound image are determined. 8.The deep learning-based ultrasound image lesion automatic identification method of claim 1, wherein, The definition, contrast and boundary sharpening degree are used to identify the sound wave arrival depth range, the echo intensity data of each layer of tissue is determined according to the contour coordinates, tissue gray distribution and internal uniformity of the lesion region, the sound wave penetration is determined according to the sound wave arrival depth range and the echo intensity data of each layer of tissue, and the ultrasound image meeting the deep imaging requirement is obtained, including: The sound wave propagation depth is estimated according to the definition and boundary sharpening degree, the maximum depth value of the sound wave arrival is determined in combination with the contrast attenuation curve; the contour coordinates and tissue gray distribution of the lesion region are used to statistically count the pixel gray value of each depth layer, and the echo intensity data set is calculated; the echo intensity attenuation rate and the sound wave penetration coefficient of adjacent layers are calculated according to the echo intensity data set, the ultrasound image meeting the deep imaging standard is screened, and the image with the highest comprehensive score of definition, contrast and boundary sharpening degree is selected.

Images