U-Net-Embedded Gabor Kernel and Coaxial Correction Methods to Dorsal Hand Vein Image Projection System
Round 1
Reviewer 1 Report
The authors present a study on the use of UNet to improve vein detection, thus assessing a clear and real medical need. The manuscript is interesting and can be accepted for publication, prior some major revision.
The described approach is based on a device and the UNet, but the illumination and camera are not correctly described: there is no indication of the producers of the light sources and their emitting parameters, nor of the camera or the projector. The dimension (in terms of pixel) of the acquired images is not mentioned. Please, deeply describe in details the experimental apparatus.
Beside this, the authors refer to Gabor transform saying that it "has biological properties": which is the exact meaning of this sentence? How can a transform have biological properties? Please clarify.
Author Response
|
Response 1: Thank you for pointing this out. We agree with this comment. In this experiment, The light source was assembled by us, so there is no manufacturer-related description in the original manuscript. We used four 850nm LED chips as the light source, connecting them in parallel to a 3.3V power supply, and measured a working current of 50mA. These four LED chips were embedded in a 5x5 centimeter acrylic light guide plate to generate uniform near-infrared light. We have included relevant descriptions of the camera and projector brands. Additionally, we have added Table 1 to the manuscript to provide detailed parameters of the camera, as shown below. Table 1. Camera parameters
In the process of capturing images, we set the image size to 192x192 pixels based on the proportion of the hand's area within the camera's captured frame. Regarding the statement about the "has biological properties", It refers to the similarity in the response of the two-dimensional Gabor filter and the biological visual system to images. Because Gabor filters exhibit frequency selectivity and orientation selectivity in processing visual information, these characteristics are similar to the biological visual system. Therefore, Gabor filters are helpful in extracting texture and structural information from images. For this section, we have rephrased the language for a clearer description. |
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The purpose of this paper is an improved U-Net for segmenting vein and a coaxial correction for image alignment in the self-build vein projection system. The manuscript is well written and provides valuable information about the study.
My overall comment is: the manuscript needs improvement. The experimental design should be presented more clearly. It is necessary to indicate the number of iterations, present the average results and the limits of parameter change. I propose to prove the increase in accuracy using statistical methods.
Author Response
Thank you for pointing this out. We agree with this comment. We have supplemented the experimental procedure, specifying the relevant parameters. Before conducting the experiment, we collected a total of 108 images of dorsal hand veins. After simple data augmentation, these images were divided into training, validation, and test sets in a ratio of 8:1:1, resulting in 432, 54, and 54 images, respectively. During the training process, we set the batch size to 4 and trained for 200 epochs, resulting in loss function curves for each model as shown in the figure.
Figure 10. Loss function curves for each model
In the final results comparison section, we trained each model five times and took the average as the final outcome. As shown in the table 4, it can be observed that the results of each model in handling the image segmentation task are quite similar. Regardless of the metric used as a standard, our proposed method consistently consumes less time and occupies less storage space while maintaining segmentation effectiveness.
Table 4. Different segmentation results of various networks comparison
|
Downsampling Network |
Miou |
Precision |
Size（M） |
Consumption time（s） |
|
VGG16 |
89.82% |
94.33% |
95 |
0.1963 |
|
ResNet50 |
88.62% |
93.02% |
168 |
0.2147 |
|
DenseNet |
90.30% |
93.66% |
111 |
0.3092 |
|
MobileNetv2 |
90.16% |
95.76% |
19 |
0.1513 |
|
Proposed method |
90.01% |
95.25 % |
15 |
0.0910 |
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors addressed the reviewers' concern. Please clearly write if the light source power is the optical power. Also cknsider that when citing a manufacturer, you should add the production site and country in brackets.
Author Response
Thank you for pointing this out. In this manuscript, We have added the optical power parameters for the near-infrared light source. When referencing the manufacturers of the camera and projector, we have added their production locations and country in the brackets, all of which are produced in Shenzhen, China.
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript has been sufficiently improved to warrant publication in Applied Sciences
Author Response
I sincerely appreciate your review of our manuscript and the valuable feedback you have provided. Your professional insights are of utmost importance to our research work, offering us valuable guidance and directions for improvement.
In your review comments, we have received valuable advice on how to enhance the quality of the paper, enabling us to further refine and develop this work. Thank you for your contributions to our research.
