Magnetic Resonance Perfusion-Weighted Imaging in Predicting Hemorrhagic Transformation of Acute Ischemic Stroke: A Retrospective Study

Round 1

Reviewer 1 Report

This retrospective study investigated the value of different thresholds of Tmax generated from perfusion-weight MRI imaging (PWI) and apparent diffusion coefficient (ADC)value in predicting hemorrhagic transformation (HT) of acute ischemic stroke. The results showed the volumes of ADC < 620X10^-6 mm²/sec and Tmax > 6s, 8s, and 10s in the HT group were all significantly larger than those in the non-HT group and were all independent risk factors for HT. Early measurement of the volume of Tmax > 10s was the highest value with a cutoff lesion volume of 10.5 ml.

Yassi et al. previously used Tmax > 14s in predicting post-stroke hemorrhagic transformation (Stroke 2013, 44:3039-43).  The authors need to discuss the reason they used the current settings and what advantages they have. 

Atrial fibrillation, NIHSS scores, and reperfusion therapy were significantly associated with HT in this study. It is better to discuss the possible mechanism.

Table 4 showed that the specificity and accuracy of Tmax > 8s were higher than what in Tmax >10s.  It is unclear why they decided to use the value measured from Tmax > 10s in predicting HT in acute ischemic stroke patients. 

The quality of the English language is acceptable, and only minor errors should be corrected. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

·       The study emphasizes the importance of the "imaging is brain" paradigm over "time is brain." Could the authors elaborate on the implications of this shift for clinical practice and patient outcomes?

·       The study found that the volume of low ADC values was positively related to HT. How does this finding compare with previous studies, and what are the potential clinical implications?

·       The paper mentions that Tmax > 10 s performed better than ADC in predicting HT. Could the authors provide more context on the significance of this finding and its potential impact on clinical decision-making?

·       The study utilized the EfficientNet B0 CNN model for feature extraction. Were other models considered, and if so, why was EfficientNet B0 chosen over them?

·       The paper mentions the use of the NCA feature selection algorithm. Could the authors elaborate on the advantages of using NCA over other feature selection methods in this context?

·       The classification results for different upper extremity regions are provided. Were there any challenges or discrepancies encountered in classifying specific regions, and how were they addressed?

·       The study mentions the potential of the proposed model as an auxiliary tool in the automatic analysis of musculoskeletal system radiographs. How does the model handle ambiguous or unclear radiographs, and what is the potential for false positives or negatives?

·       The paper cites a high accuracy rate for different regions. Were there any regions that posed particular challenges, and if so, how were these challenges addressed?

·       The study introduces a Pyramid Deep Feature Extraction model. Could the authors provide more details on the development and validation of this model?

·       The paper mentions future studies to test and enhance the model in clinical applications. Are there any specific areas or challenges that the authors foresee in the practical implementation of the model?

·       The paper does not seem to reference of MRI the recent literature from https://doi.org/10.7717/peerj-cs.1483. How does the current study's findings align or differ from the findings of this recent literature?

·       The study was conducted in accordance with the Declaration of Helsinki. Were there any ethical challenges or considerations encountered during the research, especially concerning patient data and privacy?

·       The paper mentions the use of SVM for classification. Could the authors discuss the rationale behind choosing SVM over other classification algorithms?

·       The study extracted features from images divided into patches of various sizes. How were these specific sizes determined, and were other sizes considered or tested?

·       The paper discusses the effectiveness of a deep feature extraction model. Are there any limitations or challenges associated with this model that should be considered in future research or clinical applications?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf