List of Papers Browse by Subject Areas Author List
Abstract
Reliable classification of standard abdominal ultrasound planes is essential for sonography and autonomous diagnosis. However, this task remains challenging due to the inherent low contrast and speckle noise in ultrasound imaging, further compounded by subtle inter-plane differences within subjects and substantial anatomical variations across the population. Here, we propose the Gaussian-contrastive multi-layer fusion network (GCM-Net), which comprises a learnable multi-layer feature fusion (LMFF) module to synergistically integrate hierarchical spatial details with high-level semantics, an adaptive anisotropic Gaussian refinement (AAGR) module to spatially focus on critical anatomical structures, and a supervised contrastive learning branch to enhance feature discriminability. Systematic evaluation on the publicly available FPUS23 dataset for fetal ultrasound and the private AbdoSP-9 dataset for abdominal ultrasound demonstrates that GCM-Net outperformed existing methods across key classification metrics. Additionally, GCM-Net demonstrated robust performance in real-time causal processing of clinical ultrasound videos, achieving an expert-validated accuracy of 81.25% through the temporal confidence refinement (TCR) framework. These results underscore GCM-Net’s potential not only to streamline routine clinical diagnosis but also to serve as a high-level controller for autonomous robotic ultrasound.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3367_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/diligentwaq-spec/GCM-Net
Link to the Dataset(s)
N/A
BibTex
@InProceedings{WanAnq_GCMNet_MICCAI2026,
author = { Wang, Anqi AND Li, Zihan AND Zhang, Lei AND Zhang, Xinye AND Wang, Yingni AND Yang, Xinshan AND Liu, Mingxuan AND Wang, Xiuming AND Liao, Hongen AND Ning, Guochen},
title = { { GCM-Net: Anatomy-Aware Gaussian-Contrastive Multi-layer Fusion Network for Abdominal Ultrasound Standard Plane Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16893},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces GCM-Net, an anatomy-aware Gaussian-contrastive multi-layer fusion network designed for reliable standard plane classification in abdominal ultrasound. The core contributions include a learnable multi-layer feature fusion (LMFF) module to integrate multi-scale semantic and spatial features, an adaptive anisotropic Gaussian refinement (AAGR) module that uses differentiable centroid regression and anatomy-guided spatial attention to suppress imaging artifacts, and a dual-head supervised contrastive learning framework with a memory bank to enhance feature discriminability for subtle inter-plane differences. Extensive experiments on the public FPUS23 fetal ultrasound dataset and a private AbdoSP-9 abdominal ultrasound dataset show state-of-the-art classification performance, and real-time causal video inference with temporal confidence refinement validates its clinical practicality for automated diagnosis and robotic ultrasound guidance.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper proposes an adaptive anisotropic Gaussian refinement (AAGR) module that combines differentiable anatomical centroid regression with anisotropic Gaussian attention, which is novel for standard plane classification. Unlike generic attention, this module explicitly models anatomical geometry and conforms to the ultrasound scanning sector, effectively suppressing speckle noise and irrelevant background while focusing on diagnostically critical regions. This clinically inspired design improves both accuracy and interpretability. 2.The learnable multi-layer feature fusion (LMFF) module dynamically weights and aggregates hierarchical features from a pre-trained medical foundation model. Rather than simple concatenation or summation, it learns task-adaptive weights to combine fine-grained details and high-level semantics, which is well-suited to capturing subtle anatomical differences between similar standard planes in ultrasound. 3.The dual-head classification architecture with a memory-bank-augmented supervised contrastive loss addresses key challenges in ultrasound standard plane recognition. By using historical embeddings to enlarge negative sample diversity, it significantly improves feature compactness and separability, especially for low-quality, low-contrast, and highly similar classes. The ablation study confirms this is the most impactful component. 4.The method is validated on two distinct datasets (public fetal ultrasound FPUS23 and private abdominal AbdoSP‑9 with 9 classes) and compared against multiple strong baselines including recent foundation models. Beyond static image classification, the paper further evaluates real-time causal video inference with temporal confidence refinement (TCR) and conducts an expert clinical validation, demonstrating practical utility for autonomous robotic ultrasound and clinical workflow assistance. 5.The paper provides detailed implementation settings, complete ablation results, and an anonymous public code repository, supporting transparent reproduction. The writing is well-organized, with clear motivation, methodological details, and clinical context, making the work accessible to both technical and clinical researchers.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The overall framework combines medical foundation model fine-tuning, dynamic multi-scale feature fusion, anatomical attention, and supervised contrastive learning with memory bank. While the integration is effective, the individual components and overall pipeline lack groundbreaking methodological innovation. 2.The private AbdoSP-9 dataset is collected from a single center, with no validation on multi-center, multi-device, or pathological cases. Real abdominal ultrasound involves diseases, artifacts, and patient variations that are not represented in healthy volunteers, which weakens the evidence for real-world clinical translatio 3.The ablation only validates the presence/absence of LMFF, AAGR, and SupCon, but does not analyze critical hyperparameters (e. g. , temperature τ in contrastive loss, weight λ, memory bank size, Gaussian kernel shape). The contribution of the MED-SA adapter backbone versus other pre-trained models is also not ablated, leaving unclear whether gains come from the backbone or the proposed modules. 4.Although the AAGR module is claimed to be anatomy-aware and interpretable, the paper provides no visualizations of the Gaussian attention maps, anatomical centroids, or feature distributions to support this claim. Interpretability is asserted but not demonstrated.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
No
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed GCM‑Net integrates effective components including multi‑scale feature fusion, anatomy‑aware Gaussian refinement, and memory‑augmented contrastive learning, achieving state‑of‑the‑art results on standard plane classification tasks. However, several key limitations prevent it from meeting full acceptance criteria: first, the core technical paradigm lacks sufficient novelty, as the combination of medical foundation model adaptation, anatomical attention, and supervised contrastive learning follows well‑established pipelines in medical image classification. second, clinical generalization is weakly supported due to single‑center data collection and absence of validation on pathological or multi‑device scenarios; third, the ablation study is incomplete, lacking analysis of critical hyperparameters and backbone contribution, while interpretability claims are not visually validated.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The author has addressed most concerns, so the manuscript is acceptable.
Review #2
- Please describe the contribution of the paper
This paper presents a classification model for ultrasound standard plane view localization. The authors design and cascade multiple modules to meet the requirements of real-time performance and high accuracy in ultrasound applications. They validate their approach and its differences from prior methods on both a public and a private dataset.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.From a real clinical perspective, focusing on ultrasound standard plane classification is timely. 2.GCM-Net achieves both high accuracy and meets the real-time requirements of ultrasound scanning. 3.The authors provide a realistic scenario simulation test, which strengthens the paper.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.Although the authors focus on the recognition of ultrasound standard planes, the task is narrowly formulated as a classification problem, which limits the practical applicability of the method in real clinical settings. 2.Table 1 introduces the private dataset AbdoSP-9.I note that the authors define 9 categories; however, to my knowledge, this categorization does not follow a complete clinical standard [1], and no corresponding medical justification is provided for this design. According to prior work [1], the liver alone includes 13 standard planes, suggesting that the proposed AbdoSP-9 has limited coverage. This significantly undermines the credibility of the problem setting. In addition, treating N-SP as an independent category introduces a very broad semantic space, and the manuscript does not clarify how these negative samples are constructed. These issues indicate substantial dataset-related limitations, raising concerns about the reliability of the reported metrics and their potential clinical applicability. 3.Although Table 2 reports performance on FPU23, the improvement in accuracy is marginal, and an accuracy of 99.53% may indicate severe overfitting. 4.The evaluation in Section 3.4 appears superficial. The authors should provide a more detailed description of the implementation process corresponding to Figure 2.5.The comparative analysis is insufficient. While foundation models may offer advantages during training, the authors should also report the performance of conventional frame-by-frame models on the proposed dataset to justify the necessity of GCM-Net. Furthermore, although Section 3.4 claims that GCM-Net achieves good real-time performance, quantitative metrics such as FPS are not provided for precise comparison.
[1] Zhang, J., Chen, Y., Zeng, P., Liu, Y., Diao, Y., & Liu, P. (2023). Ultra-attention: automatic recognition of liver ultrasound standard sections based on visual attention perception structures. Ultrasound in Medicine & Biology, 49(4), 1007–1017.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although this paper timely addresses the important problem of ultrasound standard plane recognition, the dataset construction lacks support from established clinical standards, and the comparative experiments are not sufficiently transparent. Therefore, I believe the current version of the paper is not suitable for publication at MICCAI.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
The rebuttal addresses several implementation-level concerns, including TCR details, latency, memory bank size, training iterations, and additional hyperparameter analysis. However, my major concern remains largely unresolved. The clinical definition and coverage of AbdoSP-9 are still not sufficiently justified as a standard abdominal ultrasound plane recognition protocol, and the NSP category remains broad and weakly specified. The validation is still limited to a single-center private dataset without external multi-device or pathological testing. Moreover, the methodological contribution mainly integrates existing components rather than introducing a substantially novel formulation.
Review #3
- Please describe the contribution of the paper
The authors present GCM-Net, an anatomy-aware framework designed for the robust classification of standard planes in abdominal ultrasound. The main contributions of paper are;
a weighted fusion module that uses learnable scalar weights to dynamically integrate hierarchical spatial details with high-level semantic features.
To handle inherent speckle noise and background artifacts, the paper introduces the AAGR module, which filters out the background/surrounding noise from features before feeding into classification head.
the authors employed a supervised contrastive loss branch alongside a first-in-first-out memory bank to increase the discriminability of morphologically similar anatomical planes.
To validate clinical utility, the authors also performed a readers study on six continuous scanning videos.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The authors proposed a system which can adapt its focus to irregular or elongated organ shapes, which is much more flexible and allows the model to be easily adapted to different types of ultrasound scans.
The proposed LMFF module that uses adjustable scalar weights to adaptively balance fine-grained spatial details with high level semantic information. Along with physical geometric constraints, such as a trapezoidal probe mask, which is effective way to filter out background noise and speckle artifacts.
The use of memory-bank-based supervised contrastive learning strategy (though not new but) allows the model to better distinguish between morphologically similar anatomical planes.
The system was validated using real-time clinical video streams and a readers study was also done to validate the key frames selected by the model which makes the claims of paper solid and clinically grounded.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
The model lacks manual ground truth for the anatomical centroids (c_x, c_y); instead, it relies on implicit supervision from the classification task, which means the attention center is optimized for prediction accuracy rather than being strictly validated against anatomical landmarks.
The size of memory bank M for contrastive loss is not provided.
TCR is credited with achieving the expert-validated clinical accuracy of 81.25%, but the authors do not provide a dedicated mathematical breakdown or a TCR subsection in the Methodology section.
Authors mention that the AbdoSP-9 dataset was curated by “chief physicians” , but the paper does not specify if multiple experts cross-verified the labels or provide a Kappa coefficient to measure agreement.
The model is described as capable of real-time processing (~15fps), yet the paper never specifies the actual processing speed (latency) of the entire pipeline on the NVIDIA RTX 4080 hardware.
For the dynamic scanning evaluation, the success rate is reported as 70.83% , but the paper does not explicitly define the mathematical threshold (e.g., a specific confidence score or time window) used to categorize a detection as a “success” vs a “failure”. Minor: There is no mention of the number of training epochs or total iterations. The authors also didn’t mention how many trainable parameter did their trainable (peft) adapters added on top of frozen encoder. The figure 1(b) should have separate boxes/outlines separating LMFF and AARG modules.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper does solve an actual problem in clinical field. The AAGR)is a clever, task-specific alternative to standard attention. The inclusion of clinical reader study shows the authors are thinking about real-world deployment, not just curated benchmarks. But the lack of explanation for TCR module and lack of anatomical ground truths for centroid regression, raises reproducibility concerns.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Thanks to the authors for the detailed rebuttal. Most of my specific concerns were directly addressed with concrete answers: 1.annotation reliability is now supported by a Fleiss’ Kappa of 0.7114 from cross-verification across one chief physician and two physicians on 30 subjects substantial agreement. 2.the TCR formulation is now fully specified (30-frame moving average, 3.0s warmup, validation-derived threshold, 60-frame patience window), which resolves the methodology gap I flagged. 3.End-to-end latency of 65.25 ms/frame on RTX 4080 (FP16/TF32), including pre- and post-processing, gives the concrete real-time figure that was missing. 4.The memory bank size (M=512), training iterations (10,000), and trainable parameter counts (~7.10M adapters, ~9.72M total) are now provided.
Two of my concerns remain unresolved: 1.The “success” vs “failure” threshold for the 70.83% dynamic scanning success rate is still not explicitly defined. This needs a one-line mathematical definition in the camera-ready. 2.The anatomical centroid claim is defended qualitatively rather than quantitatively. The authors mentioned heatmap visualizations in the appendix will help illustrate the behaviour, but MICCAI don’t allow appendix and the model still relies on implicit supervision rather than any validation against anatomical landmarks. I’d accept this as a known limitation, but it should be stated as such.
Must be included in the camera-ready: 1.full TCR formulation as described in the rebuttal. 2.kappa value and annotation cross-verification protocol in Section 3.1.3.end-to-end latency figure (65.25 ms/frame). 4./tau and /lambda ablations. 5 mathematical definition of the dynamic scanning success criterion. 6 AAGR heatmap visualizations (adjusting in Figure 2 or as separate figure) AND/OR explicit acknowledgement that anatomical centroid supervision is implicit rather than landmark-validated.
Author Feedback
We thank all reviewers for recognizing our method’s novelty, clinical relevance and realistic applicability. Below, we address the key concerns. 1.Dataset Validity(@R1,R2,R3): Generalization(R1): GCM-Net demonstrates generalizability across two dimensions: (1) multi-domain applicability, evidenced by 99.53% accuracy on the public fetal dataset FPUS23, in addition to the AbdoSP-9 dataset (2) temporal robustness, confirmed by video testing (Sec 3.4) on unseen clinical ultrasound scan with real-world patient motion. Clinical protocol(R2): AbdoSP-9 aims to provide fine-grained subxiphoid screening of multiple organs (liver, gallbladder, pancreas), including essential pancreatic planes (PB, PT) often omitted in single-organ studies, rather than expanding coverage through intercostal views as in [1]. Our 8 SPs were curated by chief physicians based on established clinical standards [2] to balance diagnostic depth with robotic navigation utility. For comparison, contemporary work like Cai et al. [3] utilizes only 6 SPs. The N-SP category includes transitional/blurred frames, specifically designed to enhance deployment robustness. Annotation verification(R3): A subset (30 subjects) was cross-verified by one chief physician and two physicians, yielding a Fleiss’ Kappa of 0.7114.Discrepancies were resolved via consensus. We will add these details to Sec 3.1.[1] Zhang J, et al. (2023). Ultra-attention: recognition of liver ultrasound standard sections. Ultrasound Med Biol, 49(4), 1007-17.[2] Zhang M (Ed). (2018). Atlas of Humman Body Ultrasound Scanning (Rev. ed.). Science Press. [3] Cai Q, et al. SEMC: Structure-Enhanced MoE Contrastive Learning. Proc. AAAI, 2026.2.TCR & Real-time Processing(@R2, R3): We will add the complete TCR formulation to the revision. The TCR applies a 30-frame moving average for temporal smoothing. Following a 3.0-second warmup, exceeding a validation-derived threshold triggers peak-tracking. To mitigate transient noise, the task locks if no new maximum is reached within a 60-frame (2.0 seconds) patience window. End-to-end latency is 65.25 ms/frame on an RTX 4080 (FP16/TF32), including pre-processing (1024x1024 resizing, normalization), inference, and post-processing, making it fully suitable for real-time clinical use. 3.AAGR & Visualizations (@R3, R1): The anatomical centroid supervision does not rely on manual annotation. Nevertheless, the regressed coordinates consistently converge to meaningful anatomical structures during optimization, demonstrating that the model can reliably localize relevant landmarks without explicit ground truth. To further support the claimed anatomy awareness, we will add qualitative visualizations in the revised appendix, including Gaussian attention heatmaps overlaid on ultrasound frames, which show that the AAGR module consistently focuses on the relevant anatomical regions. 4.Implementation Details & Ablations (@R1, R3): We will add the following implementation details in the revised manuscript. Specifically, the SAM ViT encoder is frozen. The bottleneck adapters add ~7.10M trainable parameters to the backbone. Total trainable parameters (adapters, neck, classification head) are ~9.72M. We evaluate τ and λ on AbdoSP-9.Optimal performance (81.37%) occurs at τ=0.1, where low temperature facilitates hard negative mining to distinguish subtle anatomical variances. Higher τ (e.g., 76.03% at 0.5) over-smoothes the distribution. λ=0.6 yields the peak accuracy (81.37%), balancing contrastive regularization with classification (76.49% at 0.2; 78.47% at 1.0). Notably, AAGR σ_w, σ_h is adaptively predicted, not fixed. The memory bank size M is set to 512, trained for 10000 iterations. We will expand the implementation details accordingly. The bare MED-SA backbone (72.37%) underperforms MedViTv2 (73.44%), while our GCM-Net achieves 81.37% (+9%), proving the proposed LMFF, AAGR, and SupCon drive the improvement rather than the backbone. We hope these address your concerns.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
Some reviewers show the interesting of this work. However, there are some concerns shown by the reviewers, including but not limited to:
1.Limited methodological novelty 2.Insufficient justification of task setting of classification and the private dataset. 3.The ablation study is incomplete。 4.Details are missing (e.g., TCR module, training details)
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
While concerns regarding methodological novelty remain and the contribution is best characterized as an incremental system-level innovation rather than a fundamentally new method, the reviewers generally agreed on the clinical importance of the problem, the soundness of the technical approach, and the value of the experimental validation. The work demonstrates clear practical relevance and consistent performance gains on a clinically meaningful task. Taking into account both the methodological contribution and the clinical/application impact, the paper provides sufficient merit for acceptance in my view. I therefore support acceptance. In the camera-ready version, the authors should incorporate the key clarifications from the rebuttal.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation.
The clinical definition and coverage of the private AbdoSP-9 protocol lack rigorous justification; combining multiple distinct organ systems (liver, gallbladder, pancreas) into a arbitrary 9-class paradigm without following accepted radiological standards compromises its utility as a reliable clinical benchmark. Methodologically, the network represents an incremental assembly of established components (medical foundation model adaptation, basic scalar multi-layer feature fusion, and memory-bank supervised contrastive learning) rather than a novel paradigm. Furthermore, the validation is entirely constrained to a single-center private dataset without multi-center, multi-device, or pathological variability testing, which fails to support the claims of real-world clinical translatability or robustness against genuine patient anomalies.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The paper is interesting, proposing a strong method to solve an important task. The rebuttal addressed the major reviewer concerns.
