Skin Lesion Classification Using a Deep Ensemble Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

For a technique to gain acceptance, it needs to bring “something to the table”, i.e. be able to do things that other methods cannot. This is a well written article that suffers from a number of methodological problems which results in having less imact than it potentially deserves.

1) The safe and reliable diagnosis of melanoma is, indeed, an ongoing challenge. Since melanoma is lethal, and BCC and SCC are not, there needs to be a different approach. This needs to be emphasized in the program techniques.

2) The authors cite the number of pathological lesions, but not the number of benign controls used to train the program.

3) It is not possible to deduce the number of false positives (specificity) or false negatives (sensitivity) that contributed to the accuracy in a way that has a clinical meaning. Without this, dermatologists and pathologists will not find the results of the study useful. With BCC and SCC, the false negatives (low sensitivity) can be tolerated, with melanoma they cannot.

4) The authors should divide the “precision” and “accuracy” into the false positives (specificity) and false negatives (sensitivity).

5) The authors should make some comment regarding the images that have failed the recognition criteria. Are these the ones that humans also find most challenging?

Author Response

Thank you for your precious comments.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The article presents a study on skin lesion classification leveraging a deep ensemble model composed of VGG16, Inception-V3, and ResNet-50 architectures. The research focuses on classifying skin lesions into basal cell carcinoma (BCC), squamous cell carcinoma (SCC), and melanoma using the ISIC 2018 dataset. The ensemble approach, enhanced by transfer learning and oversampling techniques, achieves a classification accuracy of 91% on the original dataset and 97% on a balanced dataset.

 

Expand Dataset Variety: Validate the model on additional datasets to ensure generalizability and robustness.

Detailed Technical Explanation: Provide more in-depth technical details on the ensemble method and model architecture.

Comparative Analysis with State-of-the-Art: Include comparisons with other cutting-edge techniques in the field to contextualize the performance of the proposed model.

Implementation Considerations: Discuss potential challenges and solutions for deploying the model in clinical settings.

 

Author Response

Thank alot for your precious comments.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Much improved manuscript, well done

Reviewer 2 Report

Comments and Suggestions for Authors

I have no more comments for this version of the paper.