Estimation of Missing DICOM Windowing Parameters in High-Dynamic-Range Radiographs Using Deep Learning
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1 The paper's focus on dynamic range compression/enhancement for improved visualization of specific regions in DICOM images is not adequately reflected in the title. The title should be revised to clearly connect the work to the field of high dynamic range (HDR) image processing [2]. HDR-related papers should be mentioned and cited.
2 DICOM image bit-depth compression inherently risks significant information loss and the introduction of artifacts, potentially impacting subsequent diagnostic accuracy. The paper needs to address how it mitigates these risks. For example, applications like bone age prediction from hand bone CT images [1], as cited, are highly sensitive to the luminance distribution within DICOM data. How does the proposed method ensure the preservation of diagnostically relevant information?
3 The methodology relies on conventional, even outdated, deep learning architectures, lacking novelty. The authors should consider incorporating more innovative approaches.
4 The experimental validation is insufficient. Comparisons with state-of-the-art (SOTA) methods are missing, and the evaluation is limited to a single dataset. The authors should expand their experiments to include multiple datasets [1] and provide comparative performance analysis against existing techniques.
[1] Coarse-to-Fine bone age regression by using multi-scale self-attention mechanism[J]. Biomedical Signal Processing and Control, 2025, 100: 107029.
[2] High dynamic range image rendering with a luminance-chromaticity independent model[C]//Intelligence Science and Big Data Engineering. Image and Video Data Engineering: 5th International Conference, IScIDE 2015, Suzhou, China, June 14-16, 2015, Revised Selected Papers, Part I 5. Springer International Publishing, 2015: 220-230.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe current manuscript has some novelty in methodology and contribution. But some revision is needed. Consider the following comments in the revised version:
- Three different deep networks are compared in the Figure 4. The performances of the deep models are dependent to the hyper-parameters. So, discuss about initial hyper-parameters of these models (ResNet-50, MobineNetV3 and VGG-16).
- It is suggested to use the phrase “pixel intensity” as “pixel value” in scientific papers.
- What is the role of “Edge detection” in your proposed approach? Discuss with details.
- The word "conclusion" is used twice in the text as section title. One of them should be deleted.
- Conclusion is too short. Discuss about the limitations and advantages of your proposed approach. Also, propose some future work ideas, if possible.
- Your proposed approach can be used widely for medical image analysis. So, it is suggested to discuss about potential applications in the introduction or related works. For example, I find two papers titled “Cervical cancer diagnosis based on modified uniform local ternary patterns and feed forward multilayer network optimized by genetic algorithm”, and titled “Cell phenotype classification using multi threshold uniform local ternary patterns in fluorescence microscope images”, which has enough relation. Cite these papers and discuss about potential application in medical diagnosis problems.
- Is it possible to generalize your proposed approach to 16-bit DICOM files? Discuss briefly.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsMost of comments have been considered by authors in this version. The revised version is better than original manuscript in terms of paper organization and technical details. The main contribution has enough novelty in definition and methodology.
