A Novel Approach in Masking of Colposcopy Images

Abstract

Masking, in the context of image processing and computer vision, refers to the process of isolating specific regions or objects of interest within an image while suppressing or removing irrelevant information. It involves the creation of a mask, which is essentially a binary image that defines which pixels in the original image belong to the region of interest and which do not. The mask typically consists of black and white pixels, where white pixels represent the areas to be preserved or highlighted, and black pixels represent the areas to be ignored or masked out. This research presents a novel approach aimed at enhancing the accuracy of masking algorithms for colposcopy images, a crucial component in the diagnosis of cervical cancer among women. While previous research has demonstrated promising outcomes in this domain, our study introduces a novel masking algorithm tailored specifically to improve segmentation accuracy. The colposcopy dataset utilized in this research is sourced from the International Agency for Research on Cancer in France.

Specification

Description:2. FIELD OF THE INVENTION:
This invention pertains to the field of cervical cancer, with a particular emphasis on masking of Colposcopy images. Masking, in the context of image processing and computer vision, refers to the process of isolating specific regions or objects of interest within an image while suppressing or removing irrelevant information. It involves the creation of a mask, which is essentially a binary image that defines which pixels in the original image belong to the region of interest and which do not. The mask typically consists of black and white pixels, where white pixels represent the areas to be preserved or highlighted, and black pixels represent the areas to be ignored or masked out. The colposcopy dataset utilized in this invention is sourced from the International Agency for Research on Cancer in France.
________________________________________
3. BACKGROUND OF THE INVENTION:
Cervical cancer remains a significant global health challenge, particularly in low-resource settings where screening and early detection methods are less accessible. Early detection is critical for improving survival rates, and one of the primary tools for diagnosing cervical abnormalities is colposcopy. Colposcopy involves the examination of the cervix using a specialized magnifying device to identify precancerous or cancerous lesions. However, the accurate interpretation of colposcopy images requires skilled expertise, and the complexity of these images can lead to variability in diagnostic accuracy. This has led to the development of automated methods in image processing and computer vision to assist clinicians in interpreting colposcopy images more accurately and efficiently. Masking, a key process in computer vision, is one such method that isolates regions of interest (such as suspicious lesions) while excluding irrelevant details, allowing clinicians to focus on the most critical parts of the image.
The invention described leverages advanced image masking techniques to improve the analysis of colposcopy images. By creating a binary mask that highlights the regions of interest (such as areas with abnormal epithelial tissue) while suppressing the background and other irrelevant information, the process can enhance the diagnostic capabilities of the images. The colposcopy dataset used in this invention comes from the International Agency for Research on Cancer (IARC) in France, a globally recognized organization that collects and curates' comprehensive data for cancer research. By utilizing this high-quality dataset, the invention aims to refine automated cervical cancer detection and diagnostic tools, ultimately reducing human error and improving outcomes in the early detection of cervical cancer.
________________________________________
4. OBJECTIVES OF THE INVENTION:
1. Develop an automated masking technique for colposcopy images to isolate specific regions of interest, such as abnormal tissue, while removing irrelevant information for enhanced analysis and interpretation.
2. Improve the accuracy and efficiency of cervical cancer diagnosis by using image processing methods that assist clinicians in identifying precancerous and cancerous lesions more effectively.
3. Utilize binary masking to create precise visual representations of regions of interest within colposcopy images, where white pixels highlight areas to be analyzed and black pixels mask out non-essential areas.
4. Leverage the colposcopy dataset from the International Agency for Research on Cancer (IARC) to develop robust and reliable image processing models that are based on high-quality, globally-recognized data.
5. Reduce diagnostic variability and human error by integrating advanced computer vision techniques into the clinical workflow, thereby improving early detection and intervention for cervical cancer.________________________________________
5. SUMMARY OF THE INVENTION:
1. The invention focuses on enhancing the diagnostic process for cervical cancer by applying advanced image processing techniques, specifically masking, to colposcopy images. Masking, in the context of this invention, involves isolating regions of interest, such as abnormal cervical tissue, while filtering out irrelevant details. By creating a binary mask, the process highlights significant areas for analysis, aiding in the detection of precancerous or cancerous lesions.
2. The colposcopy dataset used in the invention is sourced from the International Agency for Research on Cancer (IARC) in France, providing high-quality data for accurate model development. This automated masking system aims to improve diagnostic accuracy, reduce human error, and support clinicians in early detection of cervical cancer, ultimately contributing to better patient outcomes.
________________________________________

6. DETAILED DESCRIPTION OF THE INVENTION:
Step 1: Grayscale Conversion
• The input image is converted to a grayscale image using specific weightings for different color channels. This step simplifies the image by removing color information, preparing it for further processing.
Step 2: Binary Thresholding (Below Threshold)
• For each pixel in the grayscale image, if the pixel value is below or equal to a specified threshold (128), the corresponding pixel in the output binary image is set to 0 (black).
Step 3: Binary Thresholding (Above Threshold)
• For pixels in the grayscale image that exceed the threshold value, the corresponding pixels in the binary image are set to the maximum value (255, white).
Step 4: Binary Thresholding Using OpenCV
• The OpenCV function cv2.THRESH_BINARY is applied to convert the grayscale image into a binary image, setting pixels to either 0 or 255 based on the threshold value (usually around 127).
Step 5: Contour Detection
• The algorithm identifies contours in the binary image. It does so by finding the point within the contour that is furthest from the line joining the first and last points in the contour. This helps in defining the shape of the contour.
Step 6: Contour Simplification
• If the distance of a point from the contour is larger than a specified tolerance, the point is added to the simplified contour. This process is applied recursively on both sides of the maximum point to refine the contour.
Step 7: Retrieve Contours
• The OpenCV function cv2.RETR_TREE is used to retrieve all contours in the binary image. This function organizes contours hierarchically, providing information about their relationships.
Step 8: Sort Contours by Area
• The contours detected in the previous step are sorted by their area in descending order to prioritize the largest contour, which typically represents the region of interest.
Step 9: Initialize Binary Mask
• A binary mask is initialized with zeros, meaning all pixels in the matrix are set to 0, preparing for the next step where relevant areas will be marked.
Step 10: Binary Mask Assignment
• For each pixel in the binary mask matrix, the value remains 0 unless it belongs to the largest contour. If it does, the pixel value is set to 255, marking the contour.
Step 11: Draw Largest Contour
• The binary mask is used to highlight the largest contour in the image. All pixels within this contour are set to 255, creating a distinct outline of the region.
Step 12: Lesion Contour Identification
• The area within the binary mask, excluding the largest contour, is identified as the lesion contour. This step isolates the region that is most relevant for analysis.
Step 13: Draw Lesion Contour
• A function is used to fill in the lesion contour, assigning values of 0 (black) to the pixels that represent the contour.
Step 14: Invert the Binary Mask
• The binary mask is inverted so that the regions of interest are now represented by 0 (black), and the background is represented by 255 (white).
Step 15: Bounding Rectangle Calculation
• A bounding rectangle is calculated around the non-zero pixels in the inverted mask. The rectangle is defined by its top-left corner, width, and height, enclosing the region of interest.
Step 16: Adjust Bounding Box Size
• The bounding rectangle is resized by increasing its width and height by a specified factor, to ensure that the entire region of interest is captured.
Step 17: Crop the Inverted Mask
• Using the bounding rectangle, a cropped region of the inverted mask is created, containing only the pixels within the rectangle.
Step 18: Morphological Opening
• A morphological opening operation is applied to the cropped region of the inverted mask. This process removes noise and smooths the contours by using a structuring element.
Step 19: Detect Contours in Cropped Region
• Contours are detected within the cropped region using cv2.findContours. This function identifies the shapes within the cropped area of the mask.
Step 20: Filter Contours by Area
• The contours are filtered based on their area. Only contours larger than a specified minimum area are kept, ensuring that only significant regions are considered.
Step 21: Create a Region Mask
• A region mask is created based on the filtered contours. The mask is filled with the contours, marking the areas of interest.
Step 22: Extract Region of Interest
• The region of interest is extracted from the original image by applying a bitwise AND operation with the region mask. This step isolates the desired region for further analysis or visualization.
, Claims:I/We claim a method for detecting and masking regions in colposcopy images, comprising the steps of:
(a) converting the input color image into a grayscale image by applying weighted factors to the color channels;
(b) applying a thresholding technique to the grayscale image to produce a binary image, wherein pixel values below or equal to a specified threshold are set to 0, and pixel values above the threshold are set to a maximum value;
(c) identifying contours in the binary image, by detecting the contour with the maximum distance from a line connecting two points of interest within the contour;
(d) recursively simplifying the contours by incorporating the point of maximum distance and applying the process to both sides of the contour; and
(e) drawing the contour in the output image by creating a binary mask where regions of interest are highlighted and non-relevant areas are masked.2.
2. I/We claim the method of Claim 1, wherein the thresholding technique is performed using the OpenCV cv2.THRESH_BINARY function.
3. The method of Claim 1, wherein the contours are retrieved using the OpenCV cv2.RETR_TREE function to identify a hierarchical structure of contours.
4.The method of Claim 1, wherein the largest contour is determined by sorting the contours based on the area in descending order.
5. I/We claim the method of Claim 1, wherein the binary mask is created by initializing all pixel values to 0 and assigning a value of 255 to pixels within the largest contour.
6. I/We claim a method for isolating lesion contours, comprising the steps of:
(a) identifying the area within the binary mask that excludes the largest contour, referred to as the lesion contour;
(b) drawing the lesion contour by setting the binary mask values to 0 for the pixels that represent the contour.
7. I/We claim the method of Claim 6, wherein the binary mask is inverted, and a bounding rectangle is created around the region of interest based on the non-zero pixels in the inverted mask.
8. I/We claim the method of Claim 7, wherein the bounding rectangle is resized by increasing its width and height by a specified factor to ensure the entire region of interest is captured.
9. I/We claim a method for refining the masked region, comprising the steps of:
(a) applying a morphological opening operation to the cropped region of the inverted mask, using a structuring element to remove noise and smooth the contours;
(b) filtering contours based on a minimum contour area threshold, ensuring only significant regions are retained.
10. I/We claim the method of Claim 9, wherein the region of interest is extracted by applying a bitwise AND operation with the region mask to isolate the relevant area from the original image and The output binary image, generated by the method of Claim 1, wherein the image represents the masked regions of interest in a colposcopy image, highlighting lesions for diagnostic purposes.

Documents