Detecting Tumor-Associated Intracranial Hemorrhage Using Proton Magnetic Resonance Spectroscopy

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript titled "Detecting Tumor Associated Intracranial Hemorrhage Using Proton Magnetic Resonance Spectroscopy" presents a timely and relevant exploration of the use of proton magnetic resonance spectroscopy for detecting tumor-associated intracranial hemorrhage. The authors state that MRS can enhance diagnostic capabilities in cases where traditional imaging techniques may be limited. While the topic is significant, the manuscript has several weaknesses that could be addressed to enhance clarity, depth, and overall impact.

 

Improve clarity by using subheadings in the manuscript to enhance readability.

The methodology surrounding the selection of case studies and the application of 1H-MRS is not sufficiently detailed. Please include criteria for selection of the studies.

Please write about the techniques of MRS

Please include tables that summarize studies and key findings.

Adding images to the manuscript will be beneficial

Please provide a critical analysis of the limitations of MRS.

What are the the challenges of implementing MRS in practice?

Provide a more detailed discussion on recent technological advancements, including specific studies that demonstrate improved outcomes with these new techniques.

Offer specific suggestions for future studies.

 

Some references are not up to date, and the citation style is inconsistent.

Author Response

R1.1. Improve clarity by using subheadings in the manuscript to enhance readability.

A1.1. Thank you very much for your comment. We have revised each section with subheadings accordingly. For example, section “Common Clinical Practice and Limitations” now includes A. Computed tomography; B. Magnetic resonance imaging; C. Tissue biopsy; and D. Early surgery. The similar framework is applied to other sections too.

 

R1.2. The methodology surrounding the selection of case studies and the application of 1H-MRS is not sufficiently detailed. Please include criteria for selection of the studies.

A1.2. Thank you very much for your valuable comment. This paper aims to provide a scoping review of the gaps between the diagnosis and management of tumoral hemorrhage. The inclusion criteria for selecting case studies are detailed in a paragraph under the subheading “Literature Survey: Criteria for Inclusion and Exclusion.” We cited the number of papers retrieved from a PubMed search of relevant keywords to highlight the increasing trend of using MRS in brain tumor research. However, due to the paucity of research specifically on MRS for tumoral hemorrhage, a dedicated review study was not feasible. This study serves as a foundation for understanding the current challenges and limitations in this field, supporting our proposal to integrate MRS with structural MRI for the diagnosis of intracerebral hemorrhage.

In accordance with your comments, we detailed the process of selecting case studies out of the ones we read the full text (Figure 1, “Eligibility” step), as excerpt below. Articles focusing on the application of 1H-MRS for detecting brain tumors in general were not part of the scope of this perspective if the primary focus of the paper was not on reporting notable cases of tumoral hemorrhage in the brain. Rather, we aimed to provide our technical perspective on the utility of MRS in cases where tumor mass is visually obscured by hemorrhage. Additionally, we have reorganized sections of the current study to better clarify our message.

→ Page 4, Line 52: “

Literature Survey: Criteria for Inclusion and Exclusion

We focus on tumoral hemorrhages directly related to primary or metastatic brain tumors. Our scope excludes cases of hemorrhagic transformation of ischemic stroke without neoplasms in the brain, or hemorrhages occurred due to radiotherapy against tumor masses [14, 15], as these cases lack causal links to the diagnosis of brain tumors. We conducted a review of case reports written in English, which are published in 2004-2024 and available on PubMed or Web of Science. Some articles were supplemented from the other’s citations. The keywords used for searching articles in online databases were: 1) (“brain tumor” OR “brain tumour”) AND (“intracranial hemorrhage” OR “intracerebral hemorrhage” OR “intraventricular hemorrhage” OR “intratumoral hemorrhage”) AND (“diagnosis” OR “prognosis” OR “treatment”), in addition to; 2) (“brain tumor” OR “brain tumour”) AND (“intracranial hemorrhage” OR “intracerebral hemorrhage” OR “intraventricular hemorrhage” OR “intratumoral hemorrhage”) AND (“magnetic resonance imaging” OR “MR imaging” OR “MRI”) as of July 17th, 2024 (Figure 1).

In the process of selecting articles for in-depth review (Figure 1, “Eligibility” step), we aimed to include studies that reported at least one unique case of tumoral hemorrhage. Specifically, we evaluated each paper to determine whether it addressed one or more of the following points: 1) a rare case of tumoral hemorrhage involving an uncommon type of brain tumor or a highly vascularized type of lower-grade tumor; 2) an instance where advanced radiological imaging modalities (i.e., contrast-aided angiography or perfusion-weighted images, diffusion-weighted images, 1H-MRS etc.), excluding positron emission tomography, facilitated accurate diagnosis; 3) a case where the absence of advanced imaging modalities resulted in a missed or delayed diagnosis or intervention; and 4) an example where diagnostic and therapeutic efforts failed despite employing advanced imaging modalities.

Table 1 summarizes the reviewed cases based on patients’ demographics, previously known risk factors for hemorrhage, initial clinical observations, changes in clinical focus for intervention, the state of the underlying tumor, whether patients received immediate open surgery, and their survival status at the time the case was reported. We note that for studies with insufficient data on survival time, particularly for patients shown as “on treatment” in Table 1, we were limited from fully assessing the impact of diagnostic procedures and clinical interventions on patient outcomes. Table 2 presents key insights into the causes, diagnosis, and clinical interventions for tumoral hemorrhage, drawn from recent case reports published over the past five years (2019-2024).”

 

R1.3. Please write about the techniques of MRS.

A1.3. We have reorganized the paper and created a section called “Clinical Application of 1H-MRS” for clearer explanations on technical aspects of MRS. As for the techniques of MRS, please find the excerpt below.

→ Page 10, Line 185: “

1. Techniques of 1H-MRS in clinical practice

MRS is a non-invasive technique that identifies the biochemical profile of cellular composition in the brain. Specifically, 1H-MRS detects protons (1H) in metabolites that are mobile during the acquisition time and have concentrations above a detectable threshold. While other nuclei, such as carbon (13C) and phosphorus (31P), can also be targeted by MRS to provide unique insights into metabolic processes, 1H-MRS remains the most widely used due to its accessibility and compatibility with standard MR imaging equipment. The latter nuclei require specialized coils for signal acquisition, limiting their use in routine clinical practice.

The most used acquisition protocols used for 1H-MRS are point-resolved spectroscopy (PRESS [80]) and stimulated echo acquisition mode (STEAM [81]), and PRESS is often preferred over STEAM for its superior signal-to-noise ratio (SNR). Because water is the most common source of proton in the brain, a successful acquisition of 1H-MRS must perform adequate suppression of water signal. Techniques for water suppression such as chemical shift selective technique (CHESS [82]) or variable power radiofrequency pulses with optimized relaxation delays (VAPOR [83]), reduce the dominant water signal for clearer observation of metabolite signals.

In clinical practice, a single-voxel PRESS scan with water suppression focuses on a specific volume of interest (VOI) in the brain to measure concentrations of important metabolites like choline (Cho), N-acetyl aspartate (NAA), creatine (Cr), and lactate (Lac). Since the concentrations of these metabolites can vary based on scan-specific parameters and subject-specific factors such as energy metabolism levels, they are typically expressed as normalized ratios rather than absolute values (e.g., Cho/NAA, Cho/Cr).

Higher spatial resolution is necessary to distinguish distinct metabolic profiles in varying types of tissue, such as normal brain tissue, tumor peripheries, and tumor cores. This requires using smaller voxels to acquire spectra [84], but reducing voxel size decreases the SNR, leading to a decline in data quality [85]. Because of this limitation, clinical 1H-MRS typically acquires signals from a few larger VOIs, with average sizes (e.g., 2 × 2 × 2 cm³ = 8 cm³) significantly larger than those in other MR modalities. The deterioration in data quality when achieving higher spatial resolution (smaller VOI) can be partially mitigated through deep learning (DL)-aided methods, such as denoising techniques [86]. This enables finer spatial resolution tailored to brain lesions, like smaller tumors or cerebral microbleeds, facilitating detailed visualization of brain tissues without significantly increasing scan time.

For more comprehensive spatial coverage, magnetic resonance spectroscopic imaging (MRSI) is an alternative to single-voxel MRS. MRSI reconstructs spatiotemporal (two- or three-dimensional) metabolite information, allowing spectra to be acquired across an entire brain slice or volume and providing spatial advantages over single-voxel MRS. However, routine use of MRSI is hindered by drawbacks such as longer acquisition times and low reproducibility due to variations in reconstruction methods [85].

In summary, the standard clinical application of 1H-MRS involves single-voxel MRS using the PRESS sequence, typically targeting at least two VOIs: a suspected lesion and a normal counterpart. This approach is currently the most practical option for clinical routine, as it balances technical trade-offs in accessibility, signal quality, scan duration and diagnostic accuracy.”

 

R1.4. Please include tables that summarize studies and key findings.

A1.4. We have added Table 2 that includes the key points and notable insights from each study, counting the most recent 20 studies (in the recent five years, or 2019-2024).

→ Page 37: “Table 2. Summary of key findings and insights on tumoral hemorrhage, derived from a review of 20 studies (20 clinical cases) published between 2019 and 2024.”

 

R1.5. Adding images to the manuscript will be beneficial.

A1.5. Thank you for your comments. We have added Figure 3, which provides an illustrated workflow for clinicians who are utilizing MRS for identifying tumors underlying hemorrhagic lesions. This figure demonstrates how the enhanced sensitivity of MRS for detecting tumor-specific metabolites can help reduce the risk of overlooking an underlying tumor and optimize the decision-making process prior to open surgery. The figure and its description are provided below.

→ Page 40: “Figure 3. Suggested workflow compared to the current approach for utilizing MRS to detect tumor masses obscured by hemorrhage.

1. In current clinical practice, when emergency CT scans reveal intracerebral hemorrhagic lesions, structural MRI is often considered as a follow-up after clinical correlation. However, treatment decisions may rely heavily on imaging findings, potentially overlooking underlying neoplastic conditions.
2. By integrating structural MRI with MRS for cases of suspected hemorrhagic lesions observed on CT, clinicians can obtain both structural details and a biochemical profile of the lesion. We propose three volumes of interest: 1) normal tissue often positioned contralateral to the lesion (cyan), 2) suspected tumor mass (white), and 3) perilesional area (red). If the suspected lesion exhibits a spectral profile consistent with that of neoplasms and significantly differs from the profiles of normal and perilesional regions, clinicians may proceed with interventions specifically targeting the tumor mass. This suggested workflow may enhance diagnostic sensitivity and accuracy, especially when the neoplastic mass is small or obscured by hemorrhage. MRS can also provide information regarding the tumor type, which could inform biopsy strategies and further clinical decisions.”

 

R1.6. Please provide a critical analysis of the limitations of MRS.

A1.6. Please find the excerpt from the section Clinical Application of 1H-MRS below. These two points, especially the patients’ movement, are major challenges in MRS that may necessitate repeated scans,

→ Page 15, Line 297: “Despite potential utility of 1H-MRS targeting tumor underlying hemorrhage, there are some technical difficulties to overcome for its reliable use. One concern is the susceptibility artifact that causes signal loss, or spectral distortion in hemorrhagic regions, caused by blood products [111]. These distortions are not limited to hemorrhagic areas but also commonly occur in surrounding regions, including non-tumoral areas such as necrotic tissue (due to necrosis or apoptosis) and edematous regions. When tumoral mass overlaps with hemorrhagic regions, the resulting metabolomic information may become mixed unless sufficient spatial resolution is achieved to differentiate them, a challenge that may be partially addressed by employing MRSI at the cost of longer scan times. Another challenging aspect of MR imaging, particularly in brain tumor patients, is frequent patient movement. During MRS acquisition, movement effect can significantly degrade data quality, potentially compromising the accuracy required for metabolite quantification. Although excessive movement may necessitate repeated scans, some recent advancements for refining 1H-MRS spectra can improve signal reliability, and they may enable more routine use of 1H-MRS in regions prone to susceptibility artifacts. The next section discusses recent advancements in MRS processing techniques that aim to overcome technological challenges in detecting tumoral hemorrhages.”

 

R1.7. What are the challenges of implementing MRS in practice?

A1.7. Thank you for your comments. We believe that A1.6 above addresses this comment as well.

 

R1.8. Provide a more detailed discussion on recent technological advancements, including specific studies that demonstrate improved outcomes with these new techniques.

A1.8. A new section was added.

Please find the excerpt from “Enhancing Data Quality of 1H-MRS” below.

→ Page 15, Line 315 “A. Recent technological advances in MRS analysis 

One of the major advances in MRS is related to data processing than optimizing its hardware. Specifically, DL-aided approach has improved the precision of metabolites quantification [112]. This approach is more practical and feasible compared to improving the quality of the original MRS signal by increasing the magnitude of the magnetic field (from 3T to 7T), optimizing scan parameters, or adding MR modalities to accurately segment across gray matter, white matter, and cerebrospinal fluid.

Metabolite concentration in MRS signal is quantified by fitting algorithm such as nonlinear least squares fitting (NLSF) that estimates the MR spectra of given metabolites of specific concentration level. Fitting algorithms typically used in practice include LCModel [113] or QUEST [114]. They decompose complex spectra into individual metabolite signals in frequency or time domain, accounting for signal overlapping and baseline distortions. The reliability of quantitative results can be assessed by Cramér-Rao Lower Bound (CRLB), which expresses the uncertainty of each metabolite concentration as a percentage [113]; lower CRLB values indicate higher confidence in the estimated metabolite concentrations.

Recent DL-aided techniques have been developmentally focused on not only improving performance compared to traditional NLSF-based fitting algorithms but also overcoming the technical limitations presented by the CRLB value, which is the confidence indicator of NLSF (representing only the “precision” of the fitting results).

A Bayesian deep neural network model incorporating an approximated variational inference principle with Monte Carlo sampling has been developed [115-117]. This model offers statistical uncertainty that reflects levels of both accuracy and precision and achieves a lower absolute quantification error rate compared to the conventional NLSF method. By accounting for systemic errors in the model and noise in the input, it provides a comprehensive metric of precision and accuracy. This capability enables robust metabolite quantification even with lower data quality (e.g., lower SNR) depending on the diversity and quantity of training data. This advancement can potentially reduce the scan times required for the standard of spectral signal quality in MRS.”

 

R1.9. Offer specific suggestions for future studies.

A1.9. Please find the excerpt from “Enhancing Data Quality of 1H-MRS” below.

→ Page 18, Line 389: “D. Suggestions for future research directions

Future studies focusing on processing 1H-MRS data should aim to develop robust solutions for addressing limitations that reduce its diagnostic accuracy in clinical practice. Two major issues are the lower spatial resolution and signal quality of MRS, which can be particularly problematic for smaller and irregularly shaped tissues and nonhomogeneous structures with a significant amount of fluid. Improving spatial resolution could be achieved by reducing scan time, increasing the number of VOIs covering the area of interest, or mapping VOIs with customized boundary shapes. These improvements may enhance the accuracy of diagnoses and facilitate the virtual shaping of tumor masses for clearer surgical removal. In practical applications of MRS for tumoral hemorrhage, it is essential to ensure spectral quality that allows for quantification despite the presence of hemorrhage and other nonhomogeneous lesions. For precise quantification of metabolites in MRS targeting brain tumor regions, where the concentration of most metabolites is significantly degraded compared to normal structures, more strict cutoff criteria for spectral quality should be applied. Given the performance of current DL-aided methods in spectral data processing, they could be further utilized to tackle challenges in recovering and improving MRS signal quality.

The use of advanced DL-aided methods is increasing in the analysis of 1H-MRS, with growing evidence showing the clinical potential of MRS in challenging-to-diagnose pathologies such as brain tumors [108, 135, 136]. Future MRS research should strive for multisite data collection to differentiate between primary or metastatic brain tumors and blood masses or ischemic lesions in patients with ICH. This could be achieved through biochemical profiling, possibly integrating advanced DL-aided methods to enhance spatial resolution and spectral signal quality.

Increasing the accessibility of 1H-MRS would be essential for its wider adoption in hospitals, particularly those with limited access to advanced MR modalities. Recent technological advancements now allow for the easier implementation of 1H-MRS scans immediately after structural MR sessions, requiring additional scan time of around ten minutes (for two or three VOIs in single-voxel MRS). Another important aspect of DL-aided techniques is that they reduce the reliance on human expertise for data interpretation, making 1H-MRS more practical for institutions without dedicated MRS specialists. These developments have the potential to enhance the utilization of 1H-MRS in routine clinical practice.”

 

R1.10. Some references are not up to date, and the citation style is inconsistent.

A1.10. We have revised references in accordance with the guidelines of Neurology International. For Table 1, we intentionally showed the first two authors and the publication year for a clearer presentation and the ease of search for readers.

 

Reviewer 2 Report

Comments and Suggestions for Authors

This article provides a review of the use of 1H-MRS to detect intracranial hemorrhage associated with tumors. The topic of the article is important to the field of clinical neurology and neuroimaging because it explores an advanced imaging technique that may improve the diagnosis and treatment of patients with brain tumors. The focus of the study is clear and relevant to current clinical needs. However, there are several issues that require further explanation and supplementation:

1. While the review mentions the use of 1H-MRS in the detection of tumors under intracranial hemorrhage, a more in-depth exploration of the biochemical mechanisms, including changes in different metabolites in the tumor and hemorrhagic regions and their diagnostic significance, may be needed.

2. The advances in 1H-MRS techniques mentioned in the article and their use in clinical practice are valuable, but more specific details are needed on how these techniques can overcome bleeding-induced structural damage.

3. The current literature review lacks a summary of the results of relevant clinical trials. It is recommended to supplement the analysis of existing clinical trials, especially the specific application of MRS in the diagnosis of tumor hemorrhage in terms of effectiveness and clinical outcomes.

4. The article has chosen to review the literature for the period 2004-2024, which is a relatively long timeframe to cover a number of important studies and developments within the field. However, an appropriately shorter timeframe could also be considered to focus on important advances in recent years to make the review more focused on current research trends.

5. In the final part of the literature review, the article can further discuss the research gaps and future research directions in the field to provide the reader with more enlightening thoughts. For example, problems and challenges in current research can be pointed out, as well as possible solutions or research paths.

Author Response

R2.1. While the review mentions the use of 1H-MRS in the detection of tumors under intracranial hemorrhage, a more in-depth exploration of the biochemical mechanisms, including changes in different metabolites in the tumor and hemorrhagic regions and their diagnostic significance, may be needed.

A2.1. Thank you for your comments. We had some explanations on the background of biochemical profile found in 1H-MRS, and we have added specific findings of tumor itself and tumoral origin of hemorrhage in two different sections.

Please find the excerpt from “Clinical Application of 1H-MRS” below.

→ Page 12, Line 239: “B. Functionality of 1H-MRS in detecting brain tumors

Diagnostic 1H-MRS observes metabolites associated with the infiltrative growth of tumorous mass and its metabolic imbalance. As the tumor microenvironment becomes favorable, tumor cells aggressively acquire resources to sustain proliferation, leading to increased hypoxia, angiogenesis, and invasion of surrounding tissues by degrading basement membranes [7]. This process produces contrasting characteristics of tumor mass compared to the normal, which are abnormal energy metabolism and reduced cell integrity due to increased necrosis, and elevated membrane turnover in tumor-adjacent regions [96, 97]. Key findings in the tumorous tissue include: 1) elevated Lac: abnormal anaerobic metabolism increases concurrently with the concentration of Lac that is absent in normal spectra; 2) reduced NAA: more normal cells in the mass are destroyed, reflected by lower levels of neuronal integrity probed by NAA concentration; and 3) elevated Cho: as tumor cells rapidly proliferate and infiltrate the neighbor tissues, the rate of cell membrane turnover rises and the concentration of Cho, a component of cell membranes also increases [98].

A normalized Cho/NAA ratio is a particularly informative marker for brain tumor diagnosis, signaling increased membrane turnover alongside reduced cell viability [92, 96, 99, 100]. Additionally, 1H-MRS has unique sensitivity to 2-hydroxyglutarate (2-HG), a metabolite elevated exclusively in tumor cells with the oncogenic isocitrate dehydrogenase (IDH1) mutation [101, 102], frequently found in gliomas [103] and secondary glioblastomas [104]. Since IDH1 mutation is associated with prolonged survival and better response to chemotherapy [105, 106], 1H-MRS is highly informative for these subtypes of brain tumors [107, 108]. Figure 2 demonstrates how 1H-MRS identifies tumor mass that is positive for the IDH1 mutation by quantifying 2-HG.”

→ Page 13, Line 274: “C. Implementation of 1H-MRS in detecting tumoral origin of hemorrhage

Even when high-resolution MR with contrast agent is not available, 1H-MRS can detect increase in Cho/NAA or Cho/Cr that indicates higher membrane turnover related to cell proliferation, which is a marker of a tumor mass independent of hemorrhagic damage [96].”

 

R2.2. The advances in 1H-MRS techniques mentioned in the article and their use in clinical practice are valuable, but more specific details are needed on how these techniques can overcome bleeding-induced structural damage.

A2.2. Thank you for your comments.

Please find the excerpt from “Enhancing Data Quality of 1H-MRS” below.

→ Page 17, Line 361: “C. Spectral distortions

As mentioned in the previous section, spectral distortion in hemorrhagic regions is an important concern regarding data quality in 1H-MRS. Blood products in the hemorrhagic center create paramagnetic deoxyhemoglobin, which disrupts the local magnetic field and causes field inhomogeneity, leading to lower SNR and spectral distortions [111]. This results in lower-quality signals from hemorrhagic region [127], posing technological challenges in detecting tumor markers within larger hemorrhagic centers.

The quality of MR spectra in tumoral regions is also lower compared to non-neoplastic areas [128]. This difference is due to complex biological characteristics such as cell proliferation, apoptosis, and necrosis resulting from various genetic mutations and biological changes in tumor cells [96]. These changes also lead to rapid shifts in metabolic profiles, including overall degradation of brain metabolite concentration and increased lipid production [129]. Consequently, magnetic field homogeneity in tumor masses is poorer, directly impacting the quality of MR spectra acquisition, resulting in lower SNR and broader spectral linewidth [85].

Considering the data characteristics of low SNR and broad spectral linewidth in brain tumor regions with hemorrhage, recent advances in DL-aided preprocessing methods can enhance spectral signal quality acquired from hemorrhagic centers. These methods are useful in determining whether the acquired MR spectra are suitable for analysis, if the data distortion is within an acceptable range, and in restoring lower-quality data to improve the accuracy of quantifying metabolites within the tumor. For instance, DL-aided methods can identify spurious echoes [55], suppress lipid signals [56], assess the validity of quantification based on the extent of signal distortion [57, 58], and attempt to recover signals [59]. Some methods provide uncertainty information for machine-driven quantification results to measure reliability [60, 61]. This information can guide users in assessing the risks of using unstable quantification results in radiological interpretation or for research purposes. Therefore, DL-based preprocessing techniques play a crucial role in improving the quality of MR spectral data from hemorrhagic brain tumor regions, evaluating signal distortion and the validity of quantification, and restoring low-quality data to enhance the reliability of metabolite analysis. These advancements significantly improve the interpretability and safety of data for both clinical and research applications.”

 

R2.3. The current literature review lacks a summary of the results of relevant clinical trials. It is recommended to supplement the analysis of existing clinical trials, especially the specific application of MRS in the diagnosis of tumor hemorrhage in terms of effectiveness and clinical outcomes.

A2.3. We appreciate the reviewer's comment. However, 1H-MRS is not currently used for diagnosing tumoral hemorrhage and large-scale clinical trials on applying MRS for tumoral hemorrhage are not found unfortunately. This paper aims to provide a scoping review of the gaps between the diagnosis and management of tumoral hemorrhage. We intend this study serves as a foundation for understanding the current challenges and limitations in this field, supporting our proposal to integrate MRS with structural MRI for the diagnosis of intracerebral hemorrhage. Given the paucity of available data, we are unable to address the specific points raised by the reviewer at this time. Future studies with more data may be able to provide more definitive answers.

 

R2.4. The article has chosen to review the literature for the period 2004-2024, which is a relatively long timeframe to cover several important studies and developments within the field. However, an appropriately shorter timeframe could also be considered to focus on important advances in recent years to make the review more focused on current research trends.

A2.4. We would appreciate the reviewer’s kind understanding that this paper primarily aims to provide a perspective review on the gaps between diagnosis and management of tumoral hemorrhage. As suggested by the reviewer, we have incorporated a more focused analysis of recent literature (2019-2024). A table summarizing key findings and insights related to tumoral hemorrhage has been included (Table 2).

→ Page 41: “Table 2. Summary of key findings and insights on tumoral hemorrhage, derived from a review of 20 studies (20 clinical cases) published between 2019 and 2024.”

Separately, we discussed recent important technological advances and research trends of MRS under “Enhancing Data Quality of 1H-MRS”, as cited below.

→ Page 15, Line 315: “A. Recent technological advances in MRS analysis

One of the major advances in MRS is related to data processing than optimizing its hardware. Specifically, DL-aided approach has improved the precision of metabolites quantification [112]. This approach is more practical and feasible compared to improving the quality of the original MRS signal by increasing the magnitude of the magnetic field (from 3T to 7T), optimizing scan parameters, or adding MR modalities to accurately segment across gray matter, white matter, and cerebrospinal fluid.

Metabolite concentration in MRS signal is quantified by fitting algorithm such as nonlinear least squares fitting (NLSF) that estimates the MR spectra of given metabolites of specific concentration level. Fitting algorithms typically used in practice include LCModel [113] or QUEST [114]. They decompose complex spectra into individual metabolite signals in frequency or time domain, accounting for signal overlapping and baseline distortions. The reliability of quantitative results can be assessed by Cramér-Rao Lower Bound (CRLB), which expresses the uncertainty of each metabolite concentration as a percentage [113]; lower CRLB values indicate higher confidence in the estimated metabolite concentrations.

Recent DL-aided techniques have been developmentally focused on not only improving performance compared to traditional NLSF-based fitting algorithms but also overcoming the technical limitations presented by the CRLB value, which is the confidence indicator of NLSF (representing only the “precision” of the fitting results).

A Bayesian deep neural network model incorporating an approximated variational inference principle with Monte Carlo sampling has been developed [115-117]. This model offers statistical uncertainty that reflects levels of both accuracy and precision and achieves a lower absolute quantification error rate compared to the conventional NLSF method. By accounting for systemic errors in the model and noise in the input, it provides a comprehensive metric of precision and accuracy. This capability enables robust metabolite quantification even with lower data quality (e.g., lower SNR) depending on the diversity and quantity of training data. This advancement can potentially reduce the scan times required for the standard of spectral signal quality in MRS.”

 

R2.5. In the final part of the literature review, the article can further discuss the research gaps and future research directions in the field to provide the reader with more enlightening thoughts. For example, problems and challenges in current research can be pointed out, as well as possible solutions or research paths.

A2.5. As per the reviewer’s advice, a new section was added at the end of this review. Please find the excerpt below.

→ Page 18, Line 391: “D. Suggestions for future research directions

Future studies focusing on processing 1H-MRS data should aim to develop robust solutions for addressing limitations that reduce its diagnostic accuracy in clinical practice. Two major issues are the lower spatial resolution and signal quality of MRS, which can be particularly problematic for smaller and irregularly shaped tissues and nonhomogeneous structures with a significant amount of fluid. Improving spatial resolution could be achieved by reducing scan time, increasing the number of VOIs covering the area of interest, or mapping VOIs with customized boundary shapes. These improvements may enhance the accuracy of diagnoses and facilitate the virtual shaping of tumor masses for clearer surgical removal. In practical applications of MRS for tumoral hemorrhage, it is essential to ensure spectral quality that allows for quantification despite the presence of hemorrhage and other nonhomogeneous lesions. For precise quantification of metabolites in MRS targeting brain tumor regions, where the concentration of most metabolites is significantly degraded compared to normal structures, more strict cutoff criteria for spectral quality should be applied. Given the performance of current DL-aided methods in spectral data processing, they could be further utilized to tackle challenges in recovering and improving MRS signal quality.

The use of advanced DL-aided methods is increasing in the analysis of 1H-MRS, with growing evidence showing the clinical potential of MRS in challenging-to-diagnose pathologies such as brain tumors [108, 135, 136]. Future MRS research should strive for multisite data collection to differentiate between primary or metastatic brain tumors and blood masses or ischemic lesions in patients with ICH. This could be achieved through biochemical profiling, possibly integrating advanced DL-aided methods to enhance spatial resolution and spectral signal quality.

Increasing the accessibility of 1H-MRS would be essential for its wider adoption in hospitals, particularly those with limited access to advanced MR modalities. Recent technological advancements now allow for the easier implementation of 1H-MRS scans immediately after structural MR sessions, requiring additional scan time of around ten minutes (for two or three VOIs in single-voxel MRS). Another important aspect of DL-aided techniques is that they reduce the reliance on human expertise for data interpretation, making 1H-MRS more practical for institutions without dedicated MRS specialists. These developments have the potential to enhance the utilization of 1H-MRS in routine clinical practice.”

 

 

Reviewer 3 Report

Comments and Suggestions for Authors

The paper does provide valuable insights into the potential utility of 1H-MRS in detecting tumoral hemorrhage. However, there are some fundamental problems of research design, and I feel sorry to recommend reject decision. Those several key aspects are:

 

Methodological Limitations:

 

1. Case Selection Bias:

- The review only included published case reports in English from 2004-2024, potentially missing relevant cases from non-English literature

- Successful cases are more likely to be published than unsuccessful ones, potentially skewing the understanding of outcomes

- The sample size (62 cases) is relatively small for drawing robust conclusions about intervention effectiveness

 

2. Data Quality Issues:

- Many cases had incomplete information (marked as "N/A" or "N/S" in the table)

- Follow-up periods varied widely and were sometimes not specified

- Survival outcomes were inconsistently reported (some cases marked as "on treatment" without clear outcomes)

 

3. Technological Focus:

 

The paper heavily emphasizes 1H-MRS technology while potentially underexploring other diagnostic tools:

- Limited discussion of comparative effectiveness with other imaging modalities

- Insufficient analysis of cost-benefit considerations for implementing 1H-MRS

- No systematic comparison of sensitivity/specificity with other diagnostic methods

 

Analytical Weaknesses:

 

1. Statistical Analysis:

- Lacks formal statistical analysis of outcomes

- Presents raw numbers without confidence intervals or significance testing

- No systematic assessment of potential confounding variables

 

2. Causation vs. Correlation:

- The paper sometimes implies causative relationships from correlational data

- Limited control for confounding factors in analyzing survival outcomes

- Insufficient discussion of alternative explanations for observed patterns

 

3. Evidence Synthesis:

- Heavy reliance on individual case reports rather than controlled studies

- Limited integration of existing systematic reviews or meta-analyses

- No formal quality assessment of included studies

 

Clinical Application Concerns:

 

1. Practical Implementation:

- Limited discussion of resource requirements and cost implications

- Insufficient guidance on integration into existing clinical workflows

- No clear protocols for when to use 1H-MRS versus other diagnostic tools

 

2. Patient Selection:

- Unclear criteria for which patients would benefit most from 1H-MRS

- Limited discussion of contraindications or patient risk factors

- No clear guidelines for prioritizing patients when resources are limited

 

3. Generalizability:

- Cases primarily from specialized centers with advanced imaging capabilities

- May not be representative of typical clinical settings

- Limited discussion of applicability to different healthcare systems

 

All the best,

Author Response

R3.1. Case Selection Bias

R3.1.1. The review only included published case reports in English from 2004-2024, potentially missing relevant cases from non-English literature.

A3.1.1. Thank you for the suggestion. We decided to include articles written in English only, because we will not be able to critically assess the contents of the paper if it were written in other language.

 

R.3.1.2. Successful cases are more likely to be published than unsuccessful ones, potentially skewing the understanding of outcomes.

A3.1.2. We hope the reviewer understands that the papers included in this review describe various manifestations of tumoral hemorrhage cases with the clinical course, which would not be related with the assessment of the outcome of clinical trial.

 

R.3.1.3. The sample size (62 cases) is relatively small for drawing robust conclusions about intervention effectiveness.

A3.1.3. This paper provides a scoping review of current diagnostic practices for tumoral hemorrhage, not an evaluation of intervention effectiveness. We endeavored to collect as many case reports as possible by retrieving literature from 2004 to 2024, identifying 62 cases from 59 papers.

 

R3.2. Data Quality Issues

R3.2.1. Many cases had incomplete information (marked as "N/A" or "N/S" in the table).

A3.2.1. As the reviewer pointed, many studies did not specify risk factors for brain hemorrhage, which is a limitation of retrospective reviews. However, this limitation does not significantly impact our discussion of current diagnostic approaches and the clinical course of tumoral hemorrhage.

 

R.3.2.2. Follow-up periods varied widely and were sometimes not specified.

R.3.2.3. Survival outcomes were inconsistently reported (some cases marked as "on treatment" without clear outcomes).

A3.2.2. and A3.2.3. We acknowledge the reviewer's concern about the limitations of retrospective studies, which are inherent to the retrospective review study. We declared in the manuscript that insufficient data on survival time, particularly for patients “on treatment,” may limit our understanding of the impact of misdiagnosed brain tumors.

→ Page 5, Line 74: “Table 1 summarizes the reviewed cases based on patients’ demographics, previously known risk factors for hemorrhage, initial clinical observations, changes in clinical focus for intervention, the state of the underlying tumor, whether patients received immediate open surgery, and their survival status at the time the case was reported. We note that for studies with insufficient data on survival time, particularly for patients shown as “on treatment” in Table 1, we were limited from fully assessing the impact of diagnostic procedures and clinical interventions on patient outcomes.”

 

R3.3. Technological Focus

R3.3.1. The paper emphasizes 1H-MRS technology while potentially underexploring other diagnostic tools: i) Limited discussion of comparative effectiveness with other imaging modalities; ii) Insufficient analysis of cost-benefit considerations for implementing 1H-MRS; iii) No systematic comparison of sensitivity/specificity with other diagnostic methods.

A3.3.1. We appreciate the reviewer's comment. However, 1H-MRS is not currently used for diagnosing tumoral hemorrhage and large-scale clinical trials on applying MRS for tumoral hemorrhage are not found unfortunately. This paper aims to provide a scoping review of the gaps between the diagnosis and management of tumoral hemorrhage. We intend this study serves as a foundation for understanding the current challenges and limitations in this field, supporting our proposal to integrate MRS with structural MRI for the diagnosis of intracerebral hemorrhage. Given the paucity of available data, we are unable to address the specific points raised by the reviewer at this time. Future studies with more data may be able to provide more definitive answers.

R.3.3.2. Statistical Analysis: i) Lacks formal statistical analysis of outcomes; ii) Presents raw numbers without confidence intervals or significance testing; iii) No systematic assessment of potential confounding variables.

R.3.3.4. Evidence Synthesis: i) Heavy reliance on individual case reports rather than controlled studies; ii) Limited integration of existing systematic reviews or meta-analyses; iii) No formal quality assessment of included studies.

A3.3.2. and A3.3.4. This scoping review aims to identify gaps in current practices for tumor-related hemorrhage and to propose a novel diagnostic approach. As such, we deem that rigorous statistical analysis may not be necessary.

 

R.3.3.3. Causation vs. Correlation: i) The paper sometimes implies causative relationships from correlational data; ii) Limited control for confounding factors in analyzing survival outcomes; iii) Insufficient discussion of alternative explanations for observed patterns.

A3.3.3. We would like to request the reviewer to check if the comment is relevant to our paper as we did not work on the matter mentioned above.

 

R3.4. Clinical Application Concerns

R3.4.1. Practical Implementation: i) Limited discussion of resource requirements and cost implications; ii) Insufficient guidance on integration into existing clinical workflows; iii) No clear protocols for when to use 1H-MRS versus other diagnostic tools.

A3.4.1. This study focuses on identifying gaps in current practices for tumor-related hemorrhage and proposing the potential utility of 1H-MRS, rather than detailing a specific implementation process, or its quantifiable advantage compared to the other diagnostic tools. Figure 3 was added to illustrate this novel diagnostic approach to facilitate readers’ understanding.

→ Page 40: “Figure 3. Suggested workflow compared to the current approach for utilizing MRS to detect tumor masses obscured by hemorrhage.

1. In current clinical practice, when emergency CT scans reveal intracerebral hemorrhagic lesions, structural MRI is often considered as a follow-up after clinical correlation. However, treatment decisions may rely heavily on imaging findings, potentially overlooking underlying neoplastic conditions.
2. By integrating structural MRI with MRS for cases of suspected hemorrhagic lesions observed on CT, clinicians can obtain both structural details and a biochemical profile of the lesion. We propose three volumes of interest: 1) normal tissue often positioned contralateral to the lesion (cyan), 2) suspected tumor mass (white), and 3) perilesional area (red). If the suspected lesion exhibits a spectral profile consistent with that of neoplasms and significantly differs from the profiles of normal and perilesional regions, clinicians may proceed with interventions specifically targeting the tumor mass. This suggested workflow may enhance diagnostic sensitivity and accuracy, especially when the neoplastic mass is small or obscured by hemorrhage. MRS can also provide information regarding the tumor type, which could inform biopsy strategies and further clinical decisions.”

 

R.3.4.2. Patient Selection: i) Unclear criteria for which patients would benefit most from 1H-MRS; ii) Limited discussion of contraindications or patient risk factors; iii) No clear guidelines for prioritizing patients when resources are limited.

A3.4.2. Given the conceptual nature of this perspective to provide suggestions on how to improve diagnostic accuracy, specific patient selection criteria are not yet defined. We would like to suggest that patients who show signs of intracranial hemorrhage (ICH) on CT, and those without specific risk factors for ICH (e.g., hypertension or major physical injuries) or those with previous cancer history (possibility of metastatic brain tumor) may benefit the most from 1H-MRS. The contraindications and risks of 1H-MRS are like those of MRI. In resource-limited settings, patients with ICH and clinical suspicion of neoplasm should be prioritized for 1H-MRS.

→ Page 14, Line 289: “We propose that patients presenting with ICH on CT, particularly those without common risk factors for ICH (e.g., hypertension or significant physical trauma) or those with a history of cancer (raising the possibility of metastatic brain tumors), may benefit most from 1H-MRS. The contraindications and risks associated with 1H-MRS are like those of the other MR modalities. In resource-limited settings, prioritizing 1H-MRS for patients with ICH and a strong clinical suspicion of neoplasm can help optimize diagnostic and treatment strategies.”

 

R.3.4.3. Generalizability: i) Cases primarily from specialized centers with advanced imaging capabilities; ii) May not be representative of typical clinical settings; iii) Limited discussion of applicability to different healthcare systems.

A3.4.3. As this review aims to conceptualize 1H-MRS methodological approach and propose its clinical utility in tumoral hemorrhage diagnosis, a detailed discussion of generalizability, economic impact, and service delivery in different healthcare systems is beyond the scope of this paper. Future studies will be needed to address these important considerations.  

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors were very responsive to the comments. Thank you

Reviewer 3 Report

Comments and Suggestions for Authors

All comments were addressed.