A New Method for 2D-Adapted Wavelet Construction: An Application in Mass-Type Anomalies Localization in Mammographic Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper claims to propose a new method for 2-D adapted wavelet construction with an application on mass type anomalies localization in mammographic images. In general, the method seems interesting and my comments are listed as follows:

1. In Section 1, it is not recommended to provide an overly detailed description of the background. Instead, the focus should be on the topic of the application on mass type anomalies localization in mammographic images. Then, introducing the difficulties and advantages of wavelet construction and the application of this method in application.

2. In Materials and Methods section, it is suggested to use charts to illustrate the theoretical mechanism of this method and the principle that the experimental results are superior to other methods.

3. Please indicate the horizontal and vertical coordinates in the text, as there are scale values present.

4. The specific results of these methods using classical wavelets is requested to present and explain, such as data, charts, etc in the comparative experiment.

5. The authors should consider using tables to comprehensively compare the results of multiple different algorithms since the resolution of Figures 10 and 12 is too low.

6. Please provide a clear definition of the characters in the equation when they first appear, such as N1, N2, M1, M2, and other characters in Section 2.1.

 

7. The authors should attach a flow diagram of the algorithm implementation to the theoretical part to improve readability. In addition, the numbers of the images in Figures 5 and 6 are very weird, such as "20588072" and "20588334", which need explanation.

8. Other image processing methods are suggested to be discussed in the literature review part, e.g., active contour model, Active contour model based on local Kulback-Leibler divergence for fast image segmentation, Engineering Applications of Artificial Intelligence 123, 106472

Comments on the Quality of English Language

 Minor editing of English language is required.

Author Response

Please, view attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript presents a novel adaptive wavelet-based technique for the detection of mass anomalies. The methodology employs diagonal coefficients derived from the 2D Discrete Wavelet Transform, coupled with a normalized similarity measure, to effectively identify patterns. Although this technique demonstrates high sensitivity and positive predictive value in the detection of mass-like abnormalities within digital mammography images, it also exhibits limitations in specificity and negative predictive value. This leads to a notable increase in the occurrence of false negatives. The utilization of wavelet transforms in image processing, in contrast to their application in traditional signal processing, represents a forward-thinking approach to the analysis of visual data. Images, being inherently two-dimensional, provide a more intricate and richly detailed structure for analysis compared to the typically one-dimensional nature of signals. Thus, before suggesting accepting the manuscript for publication in the Applied Sciences Journal, the following point must be addressed:

1. In the introduction section, the fundamentals of wavelets are detailed, along with a brief review of the state of the art. However, the title of the paper specifies that the application is the localization of mass anomalies in mammographic images. Therefore, it is suggested that the application should also be substantiated, and a discussion should be included about works that have had the same objective but with similar methods. It is not until section 3.2 "Detection of 2-D mass-like patterns in digital mammography images" that the problem, which should have been covered in section 1, Introduction, is superficially addressed. Some works that could be considered from the point of view in studies of mass detection (morphologies and structures):

A. Oliver , et al., (2010). A review of automatic mass detection and segmentation in mammographic images. Medical Image Analysis, 14(2). https://doi.org/10.1016/j.media.2009.12.005

Medina-Ramos, et al.,  (2024). Automated Segmentation of Breast Skin for Early Cancer Diagnosis: A Multi-otsu Region Growing Approach for Detecting Skin Thickness Variations. 211–221. https://doi.org/10.1007/978-3-031-46933-6_23

D. Scutt, et al., C. P. (2014). The relationship between breast asymmetry, breast size and the occurrence of breast cancer. Http://Dx.Doi.Org/10.1259/Bjr.70.838.9404205, 70(OCT.), 1017–1021. https://doi.org/10.1259/BJR.70.838.9404205

R. Bayareh-Mancilla, et al., (2023). Automated Computer-Assisted Medical Decision-Making System Based on Morphological Shape and Skin Thickness Analysis for Asymmetry Detection in Mammographic Images. Diagnostics 2023, Vol. 13, Page 3440, 13(22), 3440. https://doi.org/10.3390/DIAGNOSTICS13223440

E.R. Price, et al., (2015). The developing asymmetry: Revisiting a perceptual and diagnostic challenge. Radiology, 274(3), 642–651. https://doi.org/10.1148/RADIOL.14132759

 

2. In line 214, clarification is required regarding the statement, “The first image of 63 × 41 contains three words whose letters are taken separately as 2-D patterns.” This description is somewhat ambiguous. Additionally, the placement of Figures 1 and 2 prior to section 3, which covers the Results, creates confusion. It would be beneficial to either reposition these figures or provide a clearer explanation in the text to ensure that the figures' relevance to the study's results is more apparent and logically structured.

3. In lines 245 and 246, “Next, five different patterns of mass-like abnormalities were extracted, particularly from INbreast images 20588072, 20588334, 22580192, 50994354, 53580858, and 21598072 (Figure 4)”, further explanation is needed regarding the rationale behind this selection. Specifically, it is not clear why these particular mass shapes were chosen over others. In reality, the selection of tumor shapes cannot be an arbitrary decision. The criteria and reasoning behind the selection of these specific mass patterns should be elaborated upon to provide clarity and justify their relevance to the study.

4. Figure 4 displays the cases in which masses were detected; however, the authors refer to these results as "patterns." This terminology is inaccurate. In line 248, the statement “Since such patterns only occur once in the corresponding images” implies that these are not, in fact, patterns. A pattern, by definition, suggests a repetitive or recurring element. In this context, since the detected features occur only once in each image, referring to them as 'masses' or 'anomalies' rather than 'patterns' would be more appropriate and accurate. The manuscript should revise this terminology to reflect the singular and unique nature of these detections in the images.

5. In lines 248 to 251, incorporating a figure to illustrate the pipeline would be highly beneficial. A visual representation, such as a flowchart or diagram, would greatly aid in understanding the process described. It would visually convey how the test set was created, the method of merging the original and pattern images, and the application of the S-LIP model. Such a figure would provide a clearer, more immediate understanding of the methodology, enhancing the readability and comprehensibility of the paper.

6. Line 254, the computing of a centroid with “skimage.measure.regionprops”, is not a Python method. It is a function of the skimage (Scikit-image) library.

7. The average IoU mean was reported as 0 with a median of 0, indicating many cases with little to no overlap between detected and actual patterns. Could you elaborate on why the average and median IoU values were so low, despite high sensitivity and positive predictive values?

8. The paper reports high sensitivity but low specificity in pattern detection. How authors interpret these results in the context of clinical usefulness, especially considering the risk of false positives in mammography?

9. Authors observed that using classical wavelets did not yield high precision compared to shapelets. Could you discuss the advantages and limitations of using shapelets over the 61 classical wavelets in this specific application?

10. Authors mention the high self-similarity of breast tissue regions, slow intensity variations, and fuzzy edges as challenges. How do these factors specifically affect the performance of your detection algorithm, and are there potential improvements that could mitigate these issues?

 

11. Given the challenges and limitations identified in this research, what are the next steps in developing and refining this detection strategy? Are there plans to incorporate other image processing or machine learning techniques to enhance detection accuracy?

Comments for author File: Comments.pdf

Comments on the Quality of English Language

none

Author Response

Please, view attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all my comments.

Comments on the Quality of English Language

Minor editing of English language is required.