Deep Learning Prediction of Axillary Lymph Node Metastasis in Breast Cancer Patients Using Clinical Implication-Applied Preprocessed CT Images
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis a well written study investigating the feasibility of using AI in detection of nodal metastases in breast cancer. The model used showed high sensitivity and specify for detecting metastatic nodal disease. This is promising and might be very helpful in the future for decision-making process and management of these patients.
Can the authors elaborate on the degree of concordance with routine standard imaging? How many of the patients with node positive disease were false negatives on initial standard radiology? Staging CT are usually obtained when the patient is high risk or suspected to have positive nodes anyway. This would suggest that these patients, or at least most of them were node positive anyway. If this is true, the value of AI will reduce give it didn’t add much to the diagnosis.
Was the effect of biological subtype of cancer studied? Lobular carcinoma is known to be radio-occult so would be interesting to how did the AI model do with these patients.
Author Response
Comments 1: Can the authors elaborate on the degree of concordance with routine standard imaging? How many of the patients with node positive disease were false negatives on initial standard radiology?
Response 1: Thank you for your constructive comments. As you suggested, comparing the predictive abilities of a experienced radiologist with our AI model using the same lymph node images would have indeed provided a more objective demonstration of our model's utility. However, our study did not include blind assessments by independent radiologists. Our team plans to conduct follow-up studies comparing the predictive accuracy of our model with that of experienced radiologists. Additionally, we aim to investigate whether our AI model designed to predict axillary lymph nodes can also demonstrate high predictive accuracy for other sites such as cervical and mediastinal lymph nodes.
Comment 2: Was the effect of biological subtype of cancer studied?
Response 2: Thank you for your interesting observation. Unfortunately, the clinical data analyzed in this study did not include information on biological subtypes. However, since lobular carcinoma is often radiooccult, researching whether our AI model performs well even in cancers with these histological characteristics presents an intriguing topic for future study.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis work studies deep-learning AI about axillary lymph node differentiation in breast cancer using Contrast-Enhanced X-ray CT before treatment management. The paper is well written, easy readable.
Only minor comments :
Introduction :
Line 55 :
Please mention that sentinel lymph node dissection is not as invasive as axillary lymph node dissection.
Line 79 :
“using radiologic images…” vs. “employing MRI and ultrasonographic images” : please be more precise about medical imaging technique you are telling about. Note that reference #9 is about Contrast Enhanced X-ray CT.
Materials and Methods :
2.2 : CT Imaging :
Line 111 :
Gantry rotation : time unit lacks.
Figure 4 : FSC cropping method and ASC cropping method pictures are inversed, are they not ?
2.3.6 : Learning the Network :
Line 191 : please write “10-3 to 10-5”.
Author Response
Comment 1: Line 55 - Please mention that sentinel lymph node dissection is not as invasive as axillary lymph node dissection.
Response 1: We have added this information to the text as suggested.
Comment 2: Line 79-“using radiologic images…” vs. “employing MRI and ultrasonographic images” : please be more precise about medical imaging technique you are telling about. Note that reference #9 is about Contrast Enhanced X-ray CT.
Response 2: I apologize for the confusion. There was an error in the order of references, which has now been corrected. Reference #9 is a review article that discusses AI models employing a variety of radiologic images, including MRI and ultrasonographic images.
[Reference 9] The Role of AI in Breast Cancer Lymph Node Classification: A Comprehensive Review. Cancers (Basel) 2023, 15, 2400.
Comment 3: Materials and Methods / 2.2 CT Imaging / Line 111 : Gantry rotation : time unit lacks.
Response 3: Thank you for the comment. We add time unit ‘s (sec)’ for gantry rotation time.
Comment 4: Figure 4 - FSC cropping method and ASC cropping method pictures are inversed, are they not ?
Response 4: You are correct. It was a simple error that the ASC and FSC methods were switched. We have corrected and replaced the figure accordingly.
Comment 5: Learning the Network - Line 191 -please write “10-3 to 10-5”
Response 5: As you suggested, this has been corrected.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThank you for replying to my queries. It’s unfortunate that there are no sufficient data to compare this to conventional imaging and estimate this new technology’s capability of detecting lobular breast cancer. These are important clinical questions and that will most likely come up in real-life practice. Making a comment about them in the discussion will reflect awareness of these issues and future direction and challenges for this technology.
Reviewer 2 Report
Comments and Suggestions for AuthorsNo other comment !
