Next Article in Journal
Experimental and Numerical Investigation of a Method for Strengthening Cold-Formed Steel Profiles in Bending
Next Article in Special Issue
A CNN CADx System for Multimodal Classification of Colorectal Polyps Combining WL, BLI, and LCI Modalities
Previous Article in Journal
Water Vapour Promotes CO2 Transport in Poly(ionic liquid)/Ionic Liquid-Based Thin-Film Composite Membranes Containing Zinc Salt for Flue Gas Treatment
Previous Article in Special Issue
Multi-Channel Feature Pyramid Networks for Prostate Segmentation, Based on Transrectal Ultrasound Imaging
 
 
Article
Peer-Review Record

A Machine Learning Approach to Predict an Early Biochemical Recurrence after a Radical Prostatectomy

Appl. Sci. 2020, 10(11), 3854; https://doi.org/10.3390/app10113854
by Seongkeun Park 1, Jieun Byun 2,* and Ji young Woo 2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3:
Appl. Sci. 2020, 10(11), 3854; https://doi.org/10.3390/app10113854
Submission received: 29 April 2020 / Revised: 27 May 2020 / Accepted: 30 May 2020 / Published: 1 June 2020
(This article belongs to the Special Issue Medical Artificial Intelligence)

Round 1

Reviewer 1 Report

The authors applied four different machine learning algorithms to the task of predicting biochemical recurrence after radical prostatectomy in a sample of 104 prostate cancer patients.

The paper is in general well-written and the research question and methodological approach are potentially interesting. However, in my opinion, the manuscript should be improved by highlighting and focusing on the work’s significance and value for future research.

In the current state of the manuscript, the merit of the work is not clear enough. If the main aim was to build a clinical prediction model, than one has to say that the performance of all evaluated models is low or even close to random (e.g., decision tree AUCs of 0.534, and 0.495).

If the main aim was to investigate the performance of a “non-classic” machine learning algorithm (i.e. auto-encoder) on a given dataset, then there should be much more focus on describing the model including its characteristics, insights gained when dealing with the specific prediction task and merit for future similar research questions. This would add much value to the manuscript. Also, there should be a comparison to a classic logistic regression model, which would probably be the first choice for the current prediction problem.

Furthermore, the following information should be discussed in the manuscript:

  • The auto-encoder was chosen because of its favorable properties when dealing with imbalanced datasets. Were there any other considerations for choosing an auto-encoder? How did class imbalances affect the other three algorithms?
  • What was the rationale for choosing the other three algorithms (decision tree, MLP, KNN) and why did the authors not include a standard logistic regression model, the most common model for such tasks?
  • Was there any missing data? If yes, how did the authors deal with missing data?

Author Response

We are very appreciate to take your precious time to review. We revised our paper to reflect your valuable advice in attached file

Reviewer 2 Report

The paper is well written. However, a major concern is the small number of samples to perform a machine learning analysis.  The authors should provide a statistical analysis of what the number of samples should be in order to ensure that the results are significant, even if such a number could only be attained in the future.  This means that the current study is a pilot study and its results may change with a higher number of samples.

Author Response

We are very appreciate to take your precious time to review. We revised our paper to reflect your valuable advice in attached file

Author Response File: Author Response.docx

Reviewer 3 Report

This paper presents a well document approach to the use of machine learning to predict an early Biochemical Recurrence after a radical prostatectomy using four well-known machine learning algorithms.  The paper does an excellent job of identifying a large set of related data elements and subsequently their significance in predicting BCR. The study group is imbalanced with 20 patients with BCR and 84 patients with non-BCR. The small numbers, all from the same setting, make it challenging to make significant meaning to the results of these experiments.  For example, the comment on line 232-233 has no significant meaning.  The authors do, however, acknowledge the limitations of this research.

The value of this paper, in my opinion, is an excellent description of the research and in the results of the comparison of the machine learning techniques.

I believe there is an error in Table 1 for age and non-BCR.

Author Response

We are very appreciate to take your precious time to review. We revised our paper to reflect your valuable advice in attached file

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have addressed my requests and where they did not, stated these limitations in the discussion section.

Back to TopTop