Exploring Deep Learning Strategies for Intervertebral Disc Herniation Detection on Veterinary MRI

Exploring Deep Learning Strategies for Intervertebral Disc Herniation Detection on Veterinary MRI | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Exploring Deep Learning Strategies for Intervertebral Disc Herniation Detection on Veterinary MRI Shoujin Huang, Guoxiong Deng, Yan Kang, Jianzhong Li, Jingyu Li, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3911180/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Intervertebral Disc Herniation (IVDH) is a common spinal disease in dogs, significantly impacting their health, mobility, and overall well-being. This study initiates an effort to automate the detection and localization of IVDH lesions in veterinary MRI scans, utilizing advanced artificial intelligence (AI) methods. A comprehensive canine IVDH dataset, comprising T2-weighted sagittal MRI images from 213 pet dogs of various breeds, ages, and sizes, was compiled and utilized to train and test the IVDH detection models. The experimental results showed that traditional two-stage detection models reliably outperformed one-stage models, including the recent You Only Look Once X (YOLOX) detector. In terms of methodology, this study introduced a novel spinal localization module, successfully integrated into different object detection models to enhance IVDH detection, achieving an average precision (AP) of up to 75.32%. Additionally, transfer learning was explored to adapt the IVDH detection model for a smaller feline dataset. Overall, this study provides insights into advancing AI for veterinary care, identifying challenges and exploring potential strategies for future development in veterinary radiology. Physical sciences/Engineering/Biomedical engineering Health sciences/Health care/Medical imaging/Magnetic resonance imaging Health sciences/Neurology/Neurological disorders/Spinal cord diseases Biological sciences/Zoology Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Intervertebral Disc Herniation (IVDH) is a common and severe spinal disease in dogs, accounting for 2.3–3.7% of veterinary hospital admissions 1 – 3 . Its manifestation in canine patients varies, depending on the type and location of the herniation. Clinical symptoms can range from mild discomfort and pain to severe neurological deficits. In more severe instances, IVDH may cause muscle atrophy and paralysis of the hind limbs, significantly impacting the quality of life of the affected dogs 2 , 4 , 5 . With the increasing availability of veterinary magnetic resonance imaging (MRI), there is an optimistic perspective for the early detection of IVDH 6 . However, a global shortage of radiologists skilled in veterinary MRI interpretations leads to diagnostic challenges 7 – 9 . While artificial intelligence (AI) or deep learning research has advanced IVDH detection in humans 10 – 14 , the anatomical differences between humans and animals, especially the smaller size of animal intervertebral discs, create unique challenges in adapting these methods for veterinary applications. Moreover, while a few studies 15 , 16 have applied AI techniques to canine IVDH, their focus has predominantly been on image quality improvement 15 or image-level disease classification 16 . The localization of IVDH lesions on segment level, as an arguably more challenging yet crucial task, remains largely unexplored yet. Addressing this gap, recent advances in deep learning-based object detection methods have paved the way. Object detection methods generally fall into two types: one-stage and two-stage detectors 17 . The two-stage methods, such as the influential Faster Region-based Convolutional Neural Network (R-CNN) method 18 , involve a two-step process. Initially, the Region Proposal Network (RPN) identifies potential regions that may contain objects. Subsequently, the detection network undertakes the tasks of classifying and accurately localizing these identified regions with bounding boxes. Although Faster R-CNN is indeed substantially “faster” than earlier two-stage methods, two-stage detection methods still generally involve more computational steps and complexity compared to one-stage methods 17 . In contrast, one-stage methods, such as the You Only Look Once (YOLO) algorithms 19 , adopt a more efficient approach. YOLO divides the image into a grid, predicting bounding boxes and class probabilities directly from this grid structure. Widely recognized for its high inference speed, this method is particularly suitable for real-time applications on mobile devices. However, one-stage methods generally trade off some accuracy, especially in detecting smaller or more irregularly shaped objects, compared to its two-stage counterparts. As one-stage detection methods continue to advance, they are increasingly seen as capable of matching the accuracy of two-stage methods 20 – 22 . However, there is still debate over whether the state-of-the-art one-stage methods can fully replace two-stage methods, especially in specialized areas 23 , 24 . For IVDH detection on veterinary MRI, where the objects of interests, i.e., discs, are small in size and the difference between normal and herniated discs is subtle, the suitability of one-stage methods remains a question. To address the above research gap, this study investigates the feasibility and methodology of AI-assisted detection of IVDH, with focus on pet dogs. Our experiments revealed that, two-stage detection models consistently outperform one-stage models in terms of IVDH detection accuracy. Furthermore, we propose a novel spinal localization module, which can robustly enhance the IVDH detection accuracy across various models. Lastly, we show that it is possible to adapt the IVDH detection model to pet cats via transfer learning, potentially broadening the applicability of the proposed method. Materials and methods Dataset compilation From September 2019 to August 2022, our study collected 487 mid-sagittal plane MRI images from 213 pet dogs. All pet owners were informed of the details of the study and signed a consent form before their dog participated in the experiments. All procedures were approved by the Ethics Committee of Shenzhen Technology University (reference number: SZTU20200208), and were carried out in accordance with relevant guidelines and regulations. All animal experiments were complied with the ARRIVE guidelines ( https://arriveguidelines.org ). The dog samples represented a variety of breeds, sexes, ages, and weights. The most frequent breeds included Poodles (n = 55), Mixed breeds (n = 31), French Bulldogs (n = 18), Pomeranians (n = 16), and Welsh Corgis (n = 13). The age range of the dogs was from 0.3 to 18 years (mean value = 5.62, standard deviation = 3.94), and their weights varied from 1.5 to 46 kg (mean value = 9.45, standard deviation = 7.83). MRI data acquisitions were performed on a super-conductive animal MRI scanner (1.5T vPetMR, GSMED) at a local veterinary hospital. A multi-slice 2D T2-weighted fast spin-echo sequence was used with the following imaging parameters: repetition time (TR) = 2895 ms, echo time (TE) = 110 ms, matrix size = 256×384, and slice thickness = 3.5 mm. For all dogs, anesthesia was induced intravenously with propofol (2.5 mg/kg), and maintained with inhaled isoflurane at a 1.5% concentration during imaging. The MRI images were initially in Digital Imaging and Communications in Medicine (DICOM) format and were later converted to .bmp files. On these images, two experienced veterinary radiologists marked the spine region and intervertebral disc herniation (IVDH) lesions using bounding boxes, using the labelMe software ( https://github.com/labelmeai/labelme ). Subsequently, the dataset was divided randomly into a training set (50%, 106 dogs) and a test set (50%, 107 dogs). Figure 1 shows the distributions of the subject weights, ages, and the number of annotations per subject. Out of the 213 dogs, 139 (64 in the training set and 75 in the test set) had at least one IVDH lesion, while 74 (42 in the training set and 32 in the test set) had none. Proposed methodology Figure 2 presents an overview of our proposed IVDH detection workflow where a coarse-to-fine strategy is employed. After obtaining annotations of the spine and IVDH lesions on MRI images, a preprocessing model, termed the spine localization module, is first trained to identify the spine regions. Once the spine regions are detected, they are cropped from the full-sized images, effectively eliminating irrelevant background tissues. The IVDH detection model (IVDH detection module) is subsequently trained on these cropped images, providing bounding boxes and confidence scores for IVDH lesions. During testing, the spine localization module first processes the raw images to identify the spine region and crop images. The cropped image is then fed into the IVDH detection module, which locates the IVDH lesions with bounding boxes and provides a confidence score for each detection. Implementation details Since the spine detection task is relatively simple, we implemented only one model, i.e., Dynamic R-CNN 25 for the spine localization module. For the IVDH detection module, our experiments involved various well-known one-stage models, including YOLOv3 26 , FCOS 27 , YOLOF 21 and YOLOX 20 , as well as various two-stage models, including faster R-CNN 18 , Cascade R-CNN 28 , Grid R-CNN 29 , Cascade Region Proposal Network (Cascade RPN) 30 , and Dynamic R-CNN 25 . These models were selected for their high impact in the field of computer vision. They encompass a broad spectrum from earlier methods such as Faster R-CNN (proposed in 2015) and YOLOv3 (proposed in 2018) to more recent approaches like Dynamic R-CNN (proposed in 2020) and YOLOX (proposed in 2021). Additionally, the YOLOX model was implemented in a small (YOLOX-S) and large (YOLOX-L) version, respectively. We used the mmdetection framework 31 to implement these models. Following standard configurations, the FCOS, YOLOF, and all two-stage models utilized ResNet50 32 backbones pretrained on ImageNet 33 , and were trained on the IVDH dataset for 100 epochs. The YOLOv3 and YOLOX models, which do not have official ResNet50 backbones, had different configurations: YOLOv3 used a DarkNet53 34 backbone pretrained on ImageNet, and was trained on the IVDH dataset for 273 epochs; YOLOX models used CSPDarkNet 35 backbones without pretraining and were trained on the IVDH dataset for 300 epochs. Note that it is a common practice not to pretrain the backbones of YOLOX 20 . For comparative purposes, we also trained IVDH detection models using a more straightforward, end-to-end approach, i.e., directly on the full-sized images. All model training and testing were conducted on an Ubuntu 22.04 server equipped with two NVIDIA RTX 3090 GPU cards. Evaluation metrics Two key evaluation metrics are employed to quantify the IVDH detection accuracy: the precision-recall (PR) curve and average precision (AP) 36 . Before defining PR curves and AP, the Intersection over Union (IoU) 36 is needed to define a true positive detection. IoU is calculated as follows: $$\begin{array}{c}IOU=\frac{area\left({B}_{p}\cap {B}_{gt}\right)}{area\left({B}_{p}\cap {B}_{gt}\right)}\#\left(1\right)\end{array}$$ where \({B}_{p}\) is the area of the predicted bounding box and \({B}_{gt}\) is the area of the ground truth bounding box. The IoU evaluates how well the predicted bounding box overlaps with the ground truth box. In this study, we used an IoU threshold of 0.5, meaning that a predicted box must have an IoU of at least 0.5 with a ground truth box to be considered a true positive detection. Given a specific IoU threshold, a PR curve can be plotted as the graphical representation of model precision (the ratio of true positive predictions to the total positive predictions) and recall (the ratio of true positive predictions to all actual positive instances) at various confidence threshold levels. Then, average precision (AP) score can be defined as the area under the PR curve: $$\begin{array}{c}AP={\int }_{0}^{1}p\left(r\right)\hspace{0.17em}dr\#\left(2\right)\end{array}$$ In Eq. (2), \(p\left(r\right)\) is the precision as a function of recall \(r\) . The integral computes the area under the curve of precision plotted against recall, from 0 to 1. In practice, since the PR curve is typically discrete, the AP score is calculated as the weighted mean of precisions achieved at each threshold, with the increase in recall from the previous threshold used as the weight 37 . Therefore, AP score is a comprehensive metric that combines the insights of both precision and recall, providing a holistic view of the model performance in object detection tasks. Transfer learning to the feline IVDH detection Considering that IVDH affects various animal species beyond dogs 1 , 38 , 39 , we pilot to extend our research to include cats. A feline IVDH dataset was constructed, consisting of 111 images from 63 cats, using the same acquisition and annotation methods as with the canine dataset. This dataset was also collected from a local veterinary hospital with written consent from the pet owners, and subsequently divided into training (n = 33) and test (n = 30) sets. With limited training samples, this feline dataset was used to explore the adaptability of models across species. We focused primarily on the Dynamic R-CNN model and evaluated four training strategies: 1) directly applying the canine model without model retraining (no retraining), 2) retraining the model on the feline dataset (retraining on cats), 3) retraining the model on a combined dataset of both dogs and cats (retraining on dogs and cats), and 4) using the canine model weights as a starting point and fine-tuning on the feline dataset (transfer learning). For strategies 1), 2), and 3), we set the learning rate at 2.5×1e − 3 . In contrast, for the transfer learning approach, the learning rate was 2.5×1e − 5 . Other model parameters remained consistent across all four methods. Results Figure 3 illustrates typical results for canine IVDH detection without the use of the spine localization module. Limited by space, only four representative methods are presented: YOLOv3, YOLOX, Faster R-CNN, and Dynamic R-CNN. These methods are chosen as they represent the earliest and most recent advancements in one-stage and two-stage methods. The first three rows show typical examples where all models effectively identify and locate most of the IVDH lesions, albeit with some occurrences of false positives or negatives. It is observed that two-stage models tend to have fewer incorrect predictions compared to one-stage models. The fourth row presents a more complex case, where the models are more likely to generate false positives outside the spinal area, emphasizing the need of developing a spine localization module. Figure 4 plots typical results of spine localization and the resulting PR curve, demonstrating the high accuracy of the trained spine localization module, achieving a notable AP of 99.8%. This enables the proposed spine localization to be a highly reliable and fully automatic step. Table 1 compares the AP scores of all tested models, both with and without the spine localization module. The results show that two-stage models outperform one-stage models irrespective of the inclusion of the spine localization module. The incorporation of this module particularly benefits Faster R-CNN and Dynamic R-CNN, with AP score increases of 5.93% and 4.18%, respectively. The positive impact of the spine localization module is further visible in the PR curves shown in Fig. 5 , where curves including the module generally surpass those without it at most recall levels. Table 1 Average precision (AP) of the trained models with and without spine localization. The two-stage models outperformed the one-stage models, and the spine localization module improved IVDH detection accuracy for nearly all models. Type Model AP with spine localization % AP without spine localization % Difference % One-stage YOLOv3 67.39 68.18 + 0.79 FCOS 66.70 69.90 + 3.20 YOLOF 60.28 60.89 + 0.61 YOLOX-s 69.72 70.91 + 1.19 YOLOX-l 66.18 71.51 + 5.33 Two-stage Faster R-CNN 67.45 73.38 + 5.93 Cascade R-CNN 69.75 74.18 + 4.43 Grid R-CNN 69.89 71.56 + 1.67 Cascade RPN 74.36 73.99 -0.37 Dynamic R-CNN 71.14 75.32 + 4.18 Figure 6 A presents a typical image with annotations from the feline dataset. Figure 6 B shows the precision-recall (PR) curves for the four training strategies evaluated on feline dataset, with the corresponding AP scores. The results highlight the difficulty in training an IVDH detection model using only the limited feline data, leading to a low AP score of 29.82%. Models trained exclusively with canine data yield satisfactory results, achieving an AP score of 63.40%, suggesting certain similarities between the anatomical structures of the two species. Interestingly, training on a combined dataset of both cats and dogs result in a slightly lower score of 60.53%. This could be due to the complexity of learning the distinct image features of both species simultaneously. In contrast, the efficacy of our transfer learning strategy is evident, leading to the highest AP score of 67.65%. This underscores the effectiveness of transfer learning in adapting models for successful application to smaller, species-specific datasets. Discussion In this study, we explored the capability of AI-assisted intervertebral disc herniation (IVDH) detection in veterinary medicine. A key development is the use of a spine localization module in the preprocessing phase. This module not only effectively removes false positives from outside the spinal area but also concentrates the model attention on the spine region. Our internal tests indicate that this approach is more effective than directly applying models to full-sized images and subsequently using a spine localization module for false positive removal. These results highlight that in medical applications, tailored preprocessing/postprocessing strategies can be essential to pursue high accuracy. Another key insight from our study is the superior performance of two-stage models. Despite the rising popularity of one-stage detectors, even the relatively old two-stage method Faster R-CNN outperforms the recent one-stage models YOLOX. The two-stage models, which first generates region proposals and then classifies these regions, is more effective for handling small target lesions. Also, the relatively slow inference speed of two-stage models is not a concern in most medical imaging applications like ours, since the imaging process itself takes minutes while running the two-stage model, even with the extra spine localization module, costs less than 1 second/subject. This finding cautions against the blanket application of the newest computer vision models to medical contexts without considering their suitability. Indeed, Fig. 7 shows that models with higher accuracy on COCO 40 , a natural image detection dataset, do not necessarily lead to higher accuracy on animal IVDH detection. The detection accuracy achieved in our study (AP score up to 75.32%), generally falls below that achieved in human IVDH, e.g., AP scores of 89.3% 14 and 92.4% 12 have been demonstrated on human lumbar disc herniation. Several factors likely contribute to this discrepancy. Firstly, the absolute size of spinal segments in dogs and cats is small. Secondly, a large imaging field-of-view (FOV) typically covering over 16 spinal segments was used in this study. A large FOV is necessary for veterinary MRI due to the varied anatomy of animals and their inability to communicate specific pain points. This, however, leads to relatively large voxel sizes, and even fewer voxels per spinal segment. On the other hand, similar human studies 10 – 13 often focus on imaging only a few spinal segments, and thus have substantially more voxels per segment. Finally, due to the absence of standardized diagnostic criteria for animal IVDH, the annotations on samples that present borderline conditions can be subjective and slightly inconsistent to certain degree, which further complicates the training of AI models. These factors together highlight the existing challenges and the importance of further research in AI-assisted veterinary medicine. The development of automated segment-level lesion localization, as in this study, is not in competition with image-level spinal disease classification 16 ; rather, it serves as a complement, providing more detailed spatial information and additional interpretability of the image-level classification results. Combining these two methods could lead to a more comprehensive diagnostic approach, where the high-level perspective of image-level classification is merged with the detailed insights from segment-level analysis. Such integration has the potential to significantly improve diagnostic accuracy and inform more precise treatment strategies. From another perspective, optimizing imaging protocols and enhancing image reconstruction quality 15 can also play a crucial role in supporting further improvements of animal IVDH diagnosis. For example, refining MRI sequences and applying advanced reconstruction algorithms for better contrast and spatial resolution could lead to clearer visualization of IVDH lesions, thereby aiding AI models in more reliable detection. In summary, this study has demonstrated the feasibility of AI-assisted intervertebral disc herniation detection for veterinary care and has explored various strategies in preprocessing, model design, and training to effectively improve detection accuracy. Our proposed strategies, supported by experimental results, offer valuable guidelines for future research. They emphasize the importance of creating methodologies that go beyond merely replicating the latest AI advancements, focusing instead on addressing the specific challenges and needs of the target domain. Declarations Acknowledgements The authors gratefully thank Ruixiang Jiang, Wenyue Xiao, and Dexing Wei for their kind technical support for data acquisition and annotation. This work was supported by Shenzhen Higher Education Stable Support Program (No. 20220716111838002), and Natural Science Foundation of Top Talent of Shenzhen Technology University (No. 20200208 and No. GDRC202134). Author Contribution SJH: Methodology, Software, Visualization, Writing - Original draft. GXD: Conceptualization, Data acquisition, Methodology, Software. YK: Data acquisition, Writing - Review & Editing. JZL: Data acquisition, Writing - Review & Editing. JYL: Methodology, Writing - Review & Editing, Funding acquisition, Supervision. MYL: Conceptualization, Writing - Original draft, Funding acquisition, Supervision. All authors reviewed the manuscript. Data availability statement The IVDH detection datasets are available from corresponding author Mengye Lyu upon reasonable request. The code for model training and weights of trained models are available at https://github.com/mylyu. Competing Interests Jianzhong Li is the founder and CEO of the company GoldenStone. Shoujin Huang, Guoxiong Deng, Yan Kang, Jingyu Li, and Mengye Lyu declare that they have no competing interests. References Da Costa, R. C., De Decker, S., Lewis, M. J., Volk, H. & Consortium (CANSORT-SCI), C. S. C. I. Diagnostic imaging in intervertebral disc disease. Frontiers in veterinary science 7 , 588338 (2020). Bergknut, N. et al. Incidence of intervertebral disk degeneration–related diseases and associated mortality rates in dogs. Journal of the American Veterinary Medical Association 240 , 1300–1309 (2012). Priester, W. A. Canine intervertebral disc disease — Occurrence by age, breed, and sex among 8,117 cases. Theriogenology 6 , 293–303 (1976). Brisson, B. A. Intervertebral disc disease in dogs. Veterinary clinics: small animal practice 40 , 829–858 (2010). Olby, N. J., da Costa, R. C., Levine, J. M., Stein, V. M. & The Canine Spinal Cord Injury Consortium (CANSORT SCI). Prognostic Factors in Canine Acute Intervertebral Disc Disease. Frontiers in Veterinary Science 7 , 596059 (2020). Besalti, O., Pekcan, Z., Sirin, Y. S. & Erbas, G. Magnetic resonance imaging findings in dogs with thoracolumbar intervertebral disk disease: 69 cases (1997–2005). Journal of the American Veterinary Medical Association 228 , 902–908 (2006). Hennessey, E., DiFazio, M., Hennessey, R. & Cassel, N. Artificial intelligence in veterinary diagnostic imaging: A literature review. Veterinary Radiology & Ultrasound 63 , 851–870 (2022). White, C., Maddox, T. W. & Mortier, J. R. Survey of factors affecting satisfaction and success of residents enrolled in European College of Veterinary Diagnostic Imaging (ECVDI) residency programs. Veterinary Radiology & Ultrasound 64 , 992–998 (2023). Owens, J. M. et al. Veterinary radiology—History, purpose, current status and future expectations. Veterinary Radiology & Ultrasound 60 , 358–362 (2019). Guinebert, S. et al. Automatic semantic segmentation and detection of vertebras and intervertebral discs by neural networks. Computer Methods and Programs in Biomedicine Update 2 , 100055 (2022). Valarmathi, G. & Devi, S. N. Automatic localization and classification of intervertebral disc herniation using hybrid classifier. Biomedical Signal Processing and Control 86 , 105291 (2023). Tsai, J.-Y. et al. Lumbar disc herniation automatic detection in magnetic resonance imaging based on deep learning. Frontiers in Bioengineering and Biotechnology 9 , 708137 (2021). Ma, S., Huang, Y., Che, X. & Gu, R. Faster RCNN-based detection of cervical spinal cord injury and disc degeneration. Journal of applied clinical medical physics 21 , 235–243 (2020). Prisilla, A. A. et al. An approach to the diagnosis of lumbar disc herniation using deep learning models. Frontiers in Bioengineering and Biotechnology 11 , 1247112 (2023). Kang, H., Noh, D., Lee, S.-K., Choi, S. & Lee, K. Deep learning-based reconstruction can improve canine thoracolumbar magnetic resonance image quality and reduce slice thickness. Veterinary Radiology & Ultrasound 64 , 1063–1070 (2023). Biercher, A. et al. Using deep learning to detect spinal cord diseases on thoracolumbar magnetic resonance images of dogs. Frontiers in veterinary science 8 , 721167 (2021). Zou, Z., Chen, K., Shi, Z., Guo, Y. & Ye, J. Object Detection in 20 Years: A Survey. Proceedings of the IEEE 111 , 257–276 (2023). Ren, S., He, K., Girshick, R. & Sun, J. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 39 , 1137–1149 (2017). Redmon, J., Divvala, S., Girshick, R. & Farhadi, A. You only look once: Unified, real-time object detection. in Proceedings of the IEEE conference on computer vision and pattern recognition 779–788 (2016). Ge, Z., Liu, S., Wang, F., Li, Z. & Sun, J. Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430 (2021). Chen, Q. et al. You only look one-level feature. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 13039–13048 (2021). Tian, Y., Pu, L. Z. C. T., Singh, R., Burt, A. D. & Carneiro, G. One-Stage Five-Class Polyp Detection and Classification. in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019) 70–73 (2019). doi:10.1109/ISBI.2019.8759521. Carranza-García, M., Torres-Mateo, J., Lara-Benítez, P. & García-Gutiérrez, J. On the Performance of One-Stage and Two-Stage Object Detectors in Autonomous Vehicles Using Camera Data. Remote Sensing 13 , 89 (2021). Xu, J., Ren, H., Cai, S. & Zhang, X. An improved faster R-CNN algorithm for assisted detection of lung nodules. Computers in Biology and Medicine 153 , 106470 (2023). Zhang, H., Chang, H., Ma, B., Wang, N. & Chen, X. Dynamic R-CNN: Towards high quality object detection via dynamic training. in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XV 16 260–275 (Springer, 2020). Redmon, J. & Farhadi, A. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767 (2018). Tian, Z., Shen, C., Chen, H. & He, T. Fcos: Fully convolutional one-stage object detection. in Proceedings of the IEEE/CVF international conference on computer vision 9627–9636 (2019). Cai, Z. & Vasconcelos, N. Cascade R-CNN: High Quality Object Detection and Instance Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence 43 , 1483–1498 (2021). Lu, X., Li, B., Yue, Y., Li, Q. & Yan, J. Grid r-cnn. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 7363–7372 (2019). Vu, T., Jang, H., Pham, T. X. & Yoo, C. D. Cascade RPN: delving into high-quality region proposal network with adaptive convolution. in Proceedings of the 33rd International Conference on Neural Information Processing Systems (Curran Associates Inc., Red Hook, NY, USA, 2019). Chen, K. et al. MMDetection: Open MMLab Detection Toolbox and Benchmark. arXiv preprint arXiv:1906.07155 (2019). He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition 770–778 (2016). Deng, J. et al. ImageNet: A large-scale hierarchical image database. in 2009 IEEE conference on computer vision and pattern recognition 248–255 (IEEE, 2009). Redmon, J. & Farhadi, A. YOLO9000: better, faster, stronger. in Proceedings of the IEEE conference on computer vision and pattern recognition 7263–7271 (2017). Bochkovskiy, A., Wang, C.-Y. & Liao, H.-Y. M. YOLOv4: Optimal Speed and Accuracy of Object Detection. Preprint at https://doi.org/10.48550/arXiv.2004.10934 (2020). Padilla, R., Netto, S. L. & da Silva, E. A. B. A Survey on Performance Metrics for Object-Detection Algorithms. in 2020 International Conference on Systems, Signals and Image Processing (IWSSIP) 237–242 (2020). doi:10.1109/IWSSIP48289.2020.9145130. Padilla, R., Passos, W. L., Dias, T. L. B., Netto, S. L. & da Silva, E. A. B. A Comparative Analysis of Object Detection Metrics with a Companion Open-Source Toolkit. Electronics 10 , 279 (2021). Farrell, M. & Fitzpatrick, N. Feline Intervertebral Disc Disease. in Advances in Intervertebral Disc Disease in Dogs and Cats 36–49 (John Wiley & Sons, Ltd, 2015). doi:10.1002/9781118940372.ch6. Amey, J. A., Liatis, T., Cherubini, G. B., De Decker, S. & Foreman, M. H. Outcomes of surgically and conservatively managed thoracolumbar and lumbosacral intervertebral disc herniations in cats. Journal of Veterinary Internal Medicine 38 , 247–257 (2024). Lin, T.-Y. et al. Microsoft COCO: Common Objects in Context. in Computer Vision – ECCV 2014 (eds. Fleet, D., Pajdla, T., Schiele, B. & Tuytelaars, T.) 740–755 (Springer International Publishing, Cham, 2014). doi:10.1007/978-3-319-10602-1_48. Additional Declarations Competing interest reported. Jianzhong Li is the founder and CEO of the company GoldenStone. Shoujin Huang, Guoxiong Deng, Yan Kang, Jingyu Li, and Mengye Lyu declare that they have no competing interests. Cite Share Download PDF Status: Published Journal Publication published 18 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 07 May, 2024 Reviews received at journal 06 May, 2024 Reviews received at journal 24 Apr, 2024 Reviewers agreed at journal 15 Apr, 2024 Reviewers agreed at journal 02 Apr, 2024 Reviewers invited by journal 01 Apr, 2024 Editor assigned by journal 22 Mar, 2024 Editor invited by journal 20 Feb, 2024 Submission checks completed at journal 20 Feb, 2024 First submitted to journal 30 Jan, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3911180","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":273819095,"identity":"57cd0fa8-2aff-4bbd-8923-c9a0990fcd8e","order_by":0,"name":"Shoujin Huang","email":"","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":false,"prefix":"","firstName":"Shoujin","middleName":"","lastName":"Huang","suffix":""},{"id":273819096,"identity":"73cf0378-da0a-4ed7-9213-d26d7db43250","order_by":1,"name":"Guoxiong Deng","email":"","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":false,"prefix":"","firstName":"Guoxiong","middleName":"","lastName":"Deng","suffix":""},{"id":273819097,"identity":"bdb39e41-84a1-4097-8baa-d087541bbbc8","order_by":2,"name":"Yan Kang","email":"","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":false,"prefix":"","firstName":"Yan","middleName":"","lastName":"Kang","suffix":""},{"id":273819098,"identity":"28eb519a-76e0-469a-b607-9018f3a3317f","order_by":3,"name":"Jianzhong Li","email":"","orcid":"","institution":"Shenzhen GoldenStone Medical Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Jianzhong","middleName":"","lastName":"Li","suffix":""},{"id":273819099,"identity":"96fc3d1b-594e-4ec2-ae8a-f91a95c7ec64","order_by":4,"name":"Jingyu Li","email":"","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":false,"prefix":"","firstName":"Jingyu","middleName":"","lastName":"Li","suffix":""},{"id":273819100,"identity":"5dcf0cd6-24a1-4fc4-b110-d0e57c20d128","order_by":5,"name":"Mengye Lyu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIiWNgGAWjYBACPhiDH0ozNhDSwgZjSDZAVJOgxeAA0VokshMfF/w6LGd8/vjzxzwMNrIbDjA/e4BfS+5m45l9h43NbiQkNvMwpBlvOMBmboBXC8/ZbdK8PYcTt91gOAjUcjhxwwEeNgkCWrb/Bmqp39x/sBGo5T8RWth7tzHz/DicYMCQzAjUcoAoLZuleRvSDWfcSGOcOccg2XjmYTYzvFr4mXk3fub5Yy3P33/8wYc3FXayfcebn+HVAgaMbc1QFiiomAmqB4E/dUQpGwWjYBSMghEKAPi4SHKaH3FpAAAAAElFTkSuQmCC","orcid":"","institution":"Shenzhen Technology University","correspondingAuthor":true,"prefix":"","firstName":"Mengye","middleName":"","lastName":"Lyu","suffix":""}],"badges":[],"createdAt":"2024-01-30 16:29:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3911180/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3911180/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-67749-5","type":"published","date":"2024-07-19T00:31:33+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":51507345,"identity":"351af46a-c465-4a4f-929f-0c296b7d4c67","added_by":"auto","created_at":"2024-02-22 19:39:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":100014,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of subject ages, weights, and annotations per subject of the compiled canine IVDH dataset. The dataset represented a variety of breeds, sexes, ages, and weights, and was divided randomly into a training set (50%, 106 dogs) and a test set (50%, 107 dogs).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/271947108e0abe197537268b.png"},{"id":51507348,"identity":"1307b641-19d6-4b23-a978-7ef58e7a0953","added_by":"auto","created_at":"2024-02-22 19:39:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":838052,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of the proposed IVDH detection workflow using deep learning models.After obtaining annotations of the spine and IVDH lesions on MRI images, the spine localization module, is trained to effectively eliminate irrelevant background tissues. The IVDH detection model (IVDH detection module) is subsequently trained on the cropped images, providing bounding boxes and confidence scores for IVDH lesions. The inference (test) phase is similarly performed with this coarse-to-fine strategy.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/ad5455a98c6493ce90a8d4d1.png"},{"id":51507347,"identity":"8fdc4a32-ba08-4d11-a1a3-f8602a21885d","added_by":"auto","created_at":"2024-02-22 19:39:18","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":837396,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of deep learning based IVDH detection without using the spine localization module. (A-C) Typical detection results, where all models could locate most of IVDH lesions annotated by the radiologists (green boxes), despite the fact that the one-stage models (YOLOv3 and YOLOX) resulted in more false negatives. (D) A more challenging case, where the models were prone to false positives outside the spinal area. A confidence score threshold of 0.4 was used for all methods.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/74490e3779ec6867d779b40e.png"},{"id":51507350,"identity":"d3a788ac-0536-408b-b7f7-d4e3f8f170ee","added_by":"auto","created_at":"2024-02-22 19:39:18","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":542770,"visible":true,"origin":"","legend":"\u003cp\u003eSpine localization results using the trained deep learning model. (A) Visualization of spine localization results on nine dogs of different body sizes. The green boxes are ground truth labeled by the radiologists, while the red boxes are detections by the deep learning model. (B) The overall precision-recall curve of spine localization. The spine localization module was found to be highly accurate as a fully automatic preprocessing step, with average precision reaching 99.8%.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/fe0d40ca9d6000fbf4241f08.png"},{"id":51507808,"identity":"f8eeb833-0fb8-4ef4-a32d-ad6fd3b1ab3e","added_by":"auto","created_at":"2024-02-22 19:47:18","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":242429,"visible":true,"origin":"","legend":"\u003cp\u003ePrecision-recall (PR) curves of the IVDH detection models with and without the spine localization module. The spine localization module led to higher detection accuracy at nearly all recall levels.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/25b935b83a820ef480fa7116.png"},{"id":51507809,"identity":"174fc400-3be2-4bc8-846b-bf8ad6042020","added_by":"auto","created_at":"2024-02-22 19:47:18","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":602701,"visible":true,"origin":"","legend":"\u003cp\u003ePilot test to adapt the IVDH detection model to feline dataset. (A) A typical image of the feline IVDH dataset. (B) Precision-recall (PR) curves obtained using four training strategies. The transfer learning strategy led to the best PR curve and highest average precision among other retraining strategies. The average precision scores are labeled in the brackets in percentage.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/a10a73b8e39701c7cd27c6e1.png"},{"id":51507349,"identity":"2f5b5d24-d704-444d-ae29-59890e65eb2b","added_by":"auto","created_at":"2024-02-22 19:39:18","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":118757,"visible":true,"origin":"","legend":"\u003cp\u003eDetection accuracy on the COCO dataset versus on the canine IVDH dataset for various models. Because the COCO dataset contains multiple classes of objects, mean average precision (mAP) is used as its accuracy metric. Higher mAP on COCO does not necessarily lead to higher average precision (AP) on IVDH detection.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/e094d3d86424fface431c4af.png"},{"id":60717403,"identity":"cb7c203a-a5cf-4c9d-b95e-cf85e1a19e68","added_by":"auto","created_at":"2024-07-20 00:31:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4125970,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3911180/v1/d82e31dd-22a1-4f76-86db-44dd5aebf02d.pdf"}],"financialInterests":"Competing interest reported. Jianzhong Li is the founder and CEO of the company GoldenStone. Shoujin Huang, Guoxiong Deng, Yan Kang, Jingyu Li, and Mengye Lyu declare that they have no competing interests.","formattedTitle":"Exploring Deep Learning Strategies for Intervertebral Disc Herniation Detection on Veterinary MRI","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIntervertebral Disc Herniation (IVDH) is a common and severe spinal disease in dogs, accounting for 2.3\u0026ndash;3.7% of veterinary hospital admissions\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Its manifestation in canine patients varies, depending on the type and location of the herniation. Clinical symptoms can range from mild discomfort and pain to severe neurological deficits. In more severe instances, IVDH may cause muscle atrophy and paralysis of the hind limbs, significantly impacting the quality of life of the affected dogs\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWith the increasing availability of veterinary magnetic resonance imaging (MRI), there is an optimistic perspective for the early detection of IVDH\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. However, a global shortage of radiologists skilled in veterinary MRI interpretations leads to diagnostic challenges\u003csup\u003e\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. While artificial intelligence (AI) or deep learning research has advanced IVDH detection in humans\u003csup\u003e\u003cspan additionalcitationids=\"CR11 CR12 CR13\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, the anatomical differences between humans and animals, especially the smaller size of animal intervertebral discs, create unique challenges in adapting these methods for veterinary applications. Moreover, while a few studies\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e have applied AI techniques to canine IVDH, their focus has predominantly been on image quality improvement\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e or image-level disease classification\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. The localization of IVDH lesions on segment level, as an arguably more challenging yet crucial task, remains largely unexplored yet.\u003c/p\u003e \u003cp\u003eAddressing this gap, recent advances in deep learning-based object detection methods have paved the way. Object detection methods generally fall into two types: one-stage and two-stage detectors\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. The two-stage methods, such as the influential Faster Region-based Convolutional Neural Network (R-CNN) method\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, involve a two-step process. Initially, the Region Proposal Network (RPN) identifies potential regions that may contain objects. Subsequently, the detection network undertakes the tasks of classifying and accurately localizing these identified regions with bounding boxes. Although Faster R-CNN is indeed substantially \u0026ldquo;faster\u0026rdquo; than earlier two-stage methods, two-stage detection methods still generally involve more computational steps and complexity compared to one-stage methods\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn contrast, one-stage methods, such as the You Only Look Once (YOLO) algorithms\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e, adopt a more efficient approach. YOLO divides the image into a grid, predicting bounding boxes and class probabilities directly from this grid structure. Widely recognized for its high inference speed, this method is particularly suitable for real-time applications on mobile devices. However, one-stage methods generally trade off some accuracy, especially in detecting smaller or more irregularly shaped objects, compared to its two-stage counterparts.\u003c/p\u003e \u003cp\u003eAs one-stage detection methods continue to advance, they are increasingly seen as capable of matching the accuracy of two-stage methods\u003csup\u003e\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. However, there is still debate over whether the state-of-the-art one-stage methods can fully replace two-stage methods, especially in specialized areas\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. For IVDH detection on veterinary MRI, where the objects of interests, i.e., discs, are small in size and the difference between normal and herniated discs is subtle, the suitability of one-stage methods remains a question.\u003c/p\u003e \u003cp\u003eTo address the above research gap, this study investigates the feasibility and methodology of AI-assisted detection of IVDH, with focus on pet dogs. Our experiments revealed that, two-stage detection models consistently outperform one-stage models in terms of IVDH detection accuracy. Furthermore, we propose a novel spinal localization module, which can robustly enhance the IVDH detection accuracy across various models. Lastly, we show that it is possible to adapt the IVDH detection model to pet cats via transfer learning, potentially broadening the applicability of the proposed method.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDataset compilation\u003c/h2\u003e \u003cp\u003eFrom September 2019 to August 2022, our study collected 487 mid-sagittal plane MRI images from 213 pet dogs. All pet owners were informed of the details of the study and signed a consent form before their dog participated in the experiments. All procedures were approved by the Ethics Committee of Shenzhen Technology University (reference number: SZTU20200208), and were carried out in accordance with relevant guidelines and regulations. All animal experiments were complied with the ARRIVE guidelines (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arriveguidelines.org\u003c/span\u003e\u003cspan address=\"https://arriveguidelines.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The dog samples represented a variety of breeds, sexes, ages, and weights. The most frequent breeds included Poodles (n\u0026thinsp;=\u0026thinsp;55), Mixed breeds (n\u0026thinsp;=\u0026thinsp;31), French Bulldogs (n\u0026thinsp;=\u0026thinsp;18), Pomeranians (n\u0026thinsp;=\u0026thinsp;16), and Welsh Corgis (n\u0026thinsp;=\u0026thinsp;13). The age range of the dogs was from 0.3 to 18 years (mean value\u0026thinsp;=\u0026thinsp;5.62, standard deviation\u0026thinsp;=\u0026thinsp;3.94), and their weights varied from 1.5 to 46 kg (mean value\u0026thinsp;=\u0026thinsp;9.45, standard deviation\u0026thinsp;=\u0026thinsp;7.83).\u003c/p\u003e \u003cp\u003eMRI data acquisitions were performed on a super-conductive animal MRI scanner (1.5T vPetMR, GSMED) at a local veterinary hospital. A multi-slice 2D T2-weighted fast spin-echo sequence was used with the following imaging parameters: repetition time (TR)\u0026thinsp;=\u0026thinsp;2895 ms, echo time (TE)\u0026thinsp;=\u0026thinsp;110 ms, matrix size\u0026thinsp;=\u0026thinsp;256\u0026times;384, and slice thickness\u0026thinsp;=\u0026thinsp;3.5 mm. For all dogs, anesthesia was induced intravenously with propofol (2.5 mg/kg), and maintained with inhaled isoflurane at a 1.5% concentration during imaging.\u003c/p\u003e \u003cp\u003eThe MRI images were initially in Digital Imaging and Communications in Medicine (DICOM) format and were later converted to .bmp files. On these images, two experienced veterinary radiologists marked the spine region and intervertebral disc herniation (IVDH) lesions using bounding boxes, using the labelMe software (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/labelmeai/labelme\u003c/span\u003e\u003cspan address=\"https://github.com/labelmeai/labelme\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Subsequently, the dataset was divided randomly into a training set (50%, 106 dogs) and a test set (50%, 107 dogs). Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the distributions of the subject weights, ages, and the number of annotations per subject. Out of the 213 dogs, 139 (64 in the training set and 75 in the test set) had at least one IVDH lesion, while 74 (42 in the training set and 32 in the test set) had none.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eProposed methodology\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents an overview of our proposed IVDH detection workflow where a coarse-to-fine strategy is employed. After obtaining annotations of the spine and IVDH lesions on MRI images, a preprocessing model, termed the spine localization module, is first trained to identify the spine regions. Once the spine regions are detected, they are cropped from the full-sized images, effectively eliminating irrelevant background tissues. The IVDH detection model (IVDH detection module) is subsequently trained on these cropped images, providing bounding boxes and confidence scores for IVDH lesions.\u003c/p\u003e \u003cp\u003eDuring testing, the spine localization module first processes the raw images to identify the spine region and crop images. The cropped image is then fed into the IVDH detection module, which locates the IVDH lesions with bounding boxes and provides a confidence score for each detection.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eImplementation details\u003c/h2\u003e \u003cp\u003eSince the spine detection task is relatively simple, we implemented only one model, i.e., Dynamic R-CNN\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e for the spine localization module.\u003c/p\u003e \u003cp\u003eFor the IVDH detection module, our experiments involved various well-known one-stage models, including YOLOv3\u003csup\u003e26\u003c/sup\u003e, FCOS\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e, YOLOF\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e and YOLOX\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e, as well as various two-stage models, including faster R-CNN\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, Cascade R-CNN\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e, Grid R-CNN\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e, Cascade Region Proposal Network (Cascade RPN)\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, and Dynamic R-CNN\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. These models were selected for their high impact in the field of computer vision. They encompass a broad spectrum from earlier methods such as Faster R-CNN (proposed in 2015) and YOLOv3 (proposed in 2018) to more recent approaches like Dynamic R-CNN (proposed in 2020) and YOLOX (proposed in 2021). Additionally, the YOLOX model was implemented in a small (YOLOX-S) and large (YOLOX-L) version, respectively.\u003c/p\u003e \u003cp\u003eWe used the mmdetection framework \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e to implement these models. Following standard configurations, the FCOS, YOLOF, and all two-stage models utilized ResNet50\u003csup\u003e32\u003c/sup\u003e backbones pretrained on ImageNet\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e, and were trained on the IVDH dataset for 100 epochs. The YOLOv3 and YOLOX models, which do not have official ResNet50 backbones, had different configurations: YOLOv3 used a DarkNet53\u003csup\u003e34\u003c/sup\u003e backbone pretrained on ImageNet, and was trained on the IVDH dataset for 273 epochs; YOLOX models used CSPDarkNet\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e backbones without pretraining and were trained on the IVDH dataset for 300 epochs. Note that it is a common practice not to pretrain the backbones of YOLOX\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. For comparative purposes, we also trained IVDH detection models using a more straightforward, end-to-end approach, i.e., directly on the full-sized images. All model training and testing were conducted on an Ubuntu 22.04 server equipped with two NVIDIA RTX 3090 GPU cards.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation metrics\u003c/h2\u003e \u003cp\u003eTwo key evaluation metrics are employed to quantify the IVDH detection accuracy: the precision-recall (PR) curve and average precision (AP)\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. Before defining PR curves and AP, the Intersection over Union (IoU)\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e is needed to define a true positive detection. IoU is calculated as follows:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\begin{array}{c}IOU=\\frac{area\\left({B}_{p}\\cap {B}_{gt}\\right)}{area\\left({B}_{p}\\cap {B}_{gt}\\right)}\\#\\left(1\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({B}_{p}\\)\u003c/span\u003e\u003c/span\u003e is the area of the predicted bounding box and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({B}_{gt}\\)\u003c/span\u003e\u003c/span\u003e is the area of the ground truth bounding box. The IoU evaluates how well the predicted bounding box overlaps with the ground truth box. In this study, we used an IoU threshold of 0.5, meaning that a predicted box must have an IoU of at least 0.5 with a ground truth box to be considered a true positive detection.\u003c/p\u003e \u003cp\u003eGiven a specific IoU threshold, a PR curve can be plotted as the graphical representation of model precision (the ratio of true positive predictions to the total positive predictions) and recall (the ratio of true positive predictions to all actual positive instances) at various confidence threshold levels. Then, average precision (AP) score can be defined as the area under the PR curve:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\begin{array}{c}AP={\\int }_{0}^{1}p\\left(r\\right)\\hspace{0.17em}dr\\#\\left(2\\right)\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn Eq.\u0026nbsp;(2), \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(p\\left(r\\right)\\)\u003c/span\u003e\u003c/span\u003e is the precision as a function of recall\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(r\\)\u003c/span\u003e\u003c/span\u003e. The integral computes the area under the curve of precision plotted against recall, from 0 to 1. In practice, since the PR curve is typically discrete, the AP score is calculated as the weighted mean of precisions achieved at each threshold, with the increase in recall from the previous threshold used as the weight\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Therefore, AP score is a comprehensive metric that combines the insights of both precision and recall, providing a holistic view of the model performance in object detection tasks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eTransfer learning to the feline IVDH detection\u003c/h2\u003e \u003cp\u003eConsidering that IVDH affects various animal species beyond dogs\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e, we pilot to extend our research to include cats. A feline IVDH dataset was constructed, consisting of 111 images from 63 cats, using the same acquisition and annotation methods as with the canine dataset. This dataset was also collected from a local veterinary hospital with written consent from the pet owners, and subsequently divided into training (n\u0026thinsp;=\u0026thinsp;33) and test (n\u0026thinsp;=\u0026thinsp;30) sets.\u003c/p\u003e \u003cp\u003eWith limited training samples, this feline dataset was used to explore the adaptability of models across species. We focused primarily on the Dynamic R-CNN model and evaluated four training strategies: 1) directly applying the canine model without model retraining (no retraining), 2) retraining the model on the feline dataset (retraining on cats), 3) retraining the model on a combined dataset of both dogs and cats (retraining on dogs and cats), and 4) using the canine model weights as a starting point and fine-tuning on the feline dataset (transfer learning). For strategies 1), 2), and 3), we set the learning rate at 2.5\u0026times;1e\u003csup\u003e\u0026minus;\u0026thinsp;3\u003c/sup\u003e. In contrast, for the transfer learning approach, the learning rate was 2.5\u0026times;1e\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e. Other model parameters remained consistent across all four methods.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates typical results for canine IVDH detection without the use of the spine localization module. Limited by space, only four representative methods are presented: YOLOv3, YOLOX, Faster R-CNN, and Dynamic R-CNN. These methods are chosen as they represent the earliest and most recent advancements in one-stage and two-stage methods. The first three rows show typical examples where all models effectively identify and locate most of the IVDH lesions, albeit with some occurrences of false positives or negatives. It is observed that two-stage models tend to have fewer incorrect predictions compared to one-stage models. The fourth row presents a more complex case, where the models are more likely to generate false positives outside the spinal area, emphasizing the need of developing a spine localization module.\u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e plots typical results of spine localization and the resulting PR curve, demonstrating the high accuracy of the trained spine localization module, achieving a notable AP of 99.8%. This enables the proposed spine localization to be a highly reliable and fully automatic step. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e compares the AP scores of all tested models, both with and without the spine localization module. The results show that two-stage models outperform one-stage models irrespective of the inclusion of the spine localization module. The incorporation of this module particularly benefits Faster R-CNN and Dynamic R-CNN, with AP score increases of 5.93% and 4.18%, respectively. The positive impact of the spine localization module is further visible in the PR curves shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, where curves including the module generally surpass those without it at most recall levels.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAverage precision (AP) of the trained models with and without spine localization. The two-stage models outperformed the one-stage models, and the spine localization module improved IVDH detection accuracy for nearly all models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eType\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAP with spine localization %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAP without spine localization %\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDifference %\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003eOne-stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYOLOv3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e67.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e68.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;0.79\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFCOS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e66.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e69.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;3.20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYOLOF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;0.61\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYOLOX-s\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e70.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;1.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYOLOX-l\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e66.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e71.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;5.33\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003eTwo-stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFaster R-CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e67.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e73.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;5.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCascade R-CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e74.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;4.43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGrid R-CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e71.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;1.67\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCascade RPN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e74.36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e73.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.37\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDynamic R-CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e71.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e75.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e+\u0026thinsp;4.18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA presents a typical image with annotations from the feline dataset. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB shows the precision-recall (PR) curves for the four training strategies evaluated on feline dataset, with the corresponding AP scores. The results highlight the difficulty in training an IVDH detection model using only the limited feline data, leading to a low AP score of 29.82%. Models trained exclusively with canine data yield satisfactory results, achieving an AP score of 63.40%, suggesting certain similarities between the anatomical structures of the two species. Interestingly, training on a combined dataset of both cats and dogs result in a slightly lower score of 60.53%. This could be due to the complexity of learning the distinct image features of both species simultaneously. In contrast, the efficacy of our transfer learning strategy is evident, leading to the highest AP score of 67.65%. This underscores the effectiveness of transfer learning in adapting models for successful application to smaller, species-specific datasets.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we explored the capability of AI-assisted intervertebral disc herniation (IVDH) detection in veterinary medicine. A key development is the use of a spine localization module in the preprocessing phase. This module not only effectively removes false positives from outside the spinal area but also concentrates the model attention on the spine region. Our internal tests indicate that this approach is more effective than directly applying models to full-sized images and subsequently using a spine localization module for false positive removal. These results highlight that in medical applications, tailored preprocessing/postprocessing strategies can be essential to pursue high accuracy.\u003c/p\u003e \u003cp\u003eAnother key insight from our study is the superior performance of two-stage models. Despite the rising popularity of one-stage detectors, even the relatively old two-stage method Faster R-CNN outperforms the recent one-stage models YOLOX. The two-stage models, which first generates region proposals and then classifies these regions, is more effective for handling small target lesions. Also, the relatively slow inference speed of two-stage models is not a concern in most medical imaging applications like ours, since the imaging process itself takes minutes while running the two-stage model, even with the extra spine localization module, costs less than 1 second/subject. This finding cautions against the blanket application of the newest computer vision models to medical contexts without considering their suitability. Indeed, Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e shows that models with higher accuracy on COCO\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e, a natural image detection dataset, do not necessarily lead to higher accuracy on animal IVDH detection.\u003c/p\u003e \u003cp\u003eThe detection accuracy achieved in our study (AP score up to 75.32%), generally falls below that achieved in human IVDH, e.g., AP scores of 89.3%\u003csup\u003e14\u003c/sup\u003e and 92.4%\u003csup\u003e12\u003c/sup\u003e have been demonstrated on human lumbar disc herniation. Several factors likely contribute to this discrepancy. Firstly, the absolute size of spinal segments in dogs and cats is small. Secondly, a large imaging field-of-view (FOV) typically covering over 16 spinal segments was used in this study. A large FOV is necessary for veterinary MRI due to the varied anatomy of animals and their inability to communicate specific pain points. This, however, leads to relatively large voxel sizes, and even fewer voxels per spinal segment. On the other hand, similar human studies\u003csup\u003e\u003cspan additionalcitationids=\"CR11 CR12\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e often focus on imaging only a few spinal segments, and thus have substantially more voxels per segment. Finally, due to the absence of standardized diagnostic criteria for animal IVDH, the annotations on samples that present borderline conditions can be subjective and slightly inconsistent to certain degree, which further complicates the training of AI models. These factors together highlight the existing challenges and the importance of further research in AI-assisted veterinary medicine.\u003c/p\u003e \u003cp\u003eThe development of automated segment-level lesion localization, as in this study, is not in competition with image-level spinal disease classification\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e; rather, it serves as a complement, providing more detailed spatial information and additional interpretability of the image-level classification results. Combining these two methods could lead to a more comprehensive diagnostic approach, where the high-level perspective of image-level classification is merged with the detailed insights from segment-level analysis. Such integration has the potential to significantly improve diagnostic accuracy and inform more precise treatment strategies. From another perspective, optimizing imaging protocols and enhancing image reconstruction quality \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e can also play a crucial role in supporting further improvements of animal IVDH diagnosis. For example, refining MRI sequences and applying advanced reconstruction algorithms for better contrast and spatial resolution could lead to clearer visualization of IVDH lesions, thereby aiding AI models in more reliable detection.\u003c/p\u003e \u003cp\u003e In summary, this study has demonstrated the feasibility of AI-assisted intervertebral disc herniation detection for veterinary care and has explored various strategies in preprocessing, model design, and training to effectively improve detection accuracy. Our proposed strategies, supported by experimental results, offer valuable guidelines for future research. They emphasize the importance of creating methodologies that go beyond merely replicating the latest AI advancements, focusing instead on addressing the specific challenges and needs of the target domain.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eThe authors gratefully thank Ruixiang Jiang, Wenyue Xiao, and Dexing Wei for their kind technical support for data acquisition and annotation.\u003c/p\u003e\n\u003cp\u003eThis work was supported by Shenzhen Higher Education Stable Support Program (No. 20220716111838002), and Natural Science Foundation of Top Talent of Shenzhen Technology University (No. 20200208 and No. GDRC202134).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eSJH: Methodology, Software, Visualization, Writing - Original draft. GXD: Conceptualization, Data acquisition, Methodology, Software. YK: Data acquisition, Writing - Review \u0026amp; Editing. JZL: Data acquisition, Writing - Review \u0026amp; Editing. JYL: Methodology, Writing - Review \u0026amp; Editing, Funding acquisition, Supervision. MYL: Conceptualization, Writing - Original draft, Funding acquisition, Supervision. All authors reviewed the manuscript.\u003c/p\u003e\n\u003ch2\u003eData availability statement\u003c/h2\u003e\n\u003cp\u003eThe IVDH detection datasets are available from corresponding author Mengye Lyu upon reasonable request. The code for model training and weights of trained models are available at https://github.com/mylyu.\u003c/p\u003e\n\u003ch2\u003eCompeting Interests\u003c/h2\u003e\n\u003cp\u003eJianzhong Li is the founder and CEO of the company GoldenStone. Shoujin Huang, Guoxiong Deng, Yan Kang, Jingyu Li, and Mengye Lyu declare that they have no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDa Costa, R. C., De Decker, S., Lewis, M. J., Volk, H. \u0026amp; Consortium (CANSORT-SCI), C. S. C. I. Diagnostic imaging in intervertebral disc disease. \u003cem\u003eFrontiers in veterinary science\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 588338 (2020).\u003c/li\u003e\n\u003cli\u003eBergknut, N. \u003cem\u003eet al.\u003c/em\u003e Incidence of intervertebral disk degeneration\u0026ndash;related diseases and associated mortality rates in dogs. \u003cem\u003eJournal of the American Veterinary Medical Association\u003c/em\u003e \u003cstrong\u003e240\u003c/strong\u003e, 1300\u0026ndash;1309 (2012).\u003c/li\u003e\n\u003cli\u003ePriester, W. A. Canine intervertebral disc disease \u0026mdash; Occurrence by age, breed, and sex among 8,117 cases. \u003cem\u003eTheriogenology\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 293\u0026ndash;303 (1976).\u003c/li\u003e\n\u003cli\u003eBrisson, B. A. Intervertebral disc disease in dogs. \u003cem\u003eVeterinary clinics: small animal practice\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 829\u0026ndash;858 (2010).\u003c/li\u003e\n\u003cli\u003eOlby, N. J., da Costa, R. C., Levine, J. M., Stein, V. M. \u0026amp; The Canine Spinal Cord Injury Consortium (CANSORT SCI). Prognostic Factors in Canine Acute Intervertebral Disc Disease. \u003cem\u003eFrontiers in Veterinary Science\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 596059 (2020).\u003c/li\u003e\n\u003cli\u003eBesalti, O., Pekcan, Z., Sirin, Y. S. \u0026amp; Erbas, G. Magnetic resonance imaging findings in dogs with thoracolumbar intervertebral disk disease: 69 cases (1997\u0026ndash;2005). \u003cem\u003eJournal of the American Veterinary Medical Association\u003c/em\u003e \u003cstrong\u003e228\u003c/strong\u003e, 902\u0026ndash;908 (2006).\u003c/li\u003e\n\u003cli\u003eHennessey, E., DiFazio, M., Hennessey, R. \u0026amp; Cassel, N. Artificial intelligence in veterinary diagnostic imaging: A literature review. \u003cem\u003eVeterinary Radiology \u0026amp; Ultrasound\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 851\u0026ndash;870 (2022).\u003c/li\u003e\n\u003cli\u003eWhite, C., Maddox, T. W. \u0026amp; Mortier, J. R. Survey of factors affecting satisfaction and success of residents enrolled in European College of Veterinary Diagnostic Imaging (ECVDI) residency programs. \u003cem\u003eVeterinary Radiology \u0026amp; Ultrasound\u003c/em\u003e \u003cstrong\u003e64\u003c/strong\u003e, 992\u0026ndash;998 (2023).\u003c/li\u003e\n\u003cli\u003eOwens, J. M. \u003cem\u003eet al.\u003c/em\u003e Veterinary radiology\u0026mdash;History, purpose, current status and future expectations. \u003cem\u003eVeterinary Radiology \u0026amp; Ultrasound\u003c/em\u003e \u003cstrong\u003e60\u003c/strong\u003e, 358\u0026ndash;362 (2019).\u003c/li\u003e\n\u003cli\u003eGuinebert, S. \u003cem\u003eet al.\u003c/em\u003e Automatic semantic segmentation and detection of vertebras and intervertebral discs by neural networks. \u003cem\u003eComputer Methods and Programs in Biomedicine Update\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 100055 (2022).\u003c/li\u003e\n\u003cli\u003eValarmathi, G. \u0026amp; Devi, S. N. Automatic localization and classification of intervertebral disc herniation using hybrid classifier. \u003cem\u003eBiomedical Signal Processing and Control\u003c/em\u003e \u003cstrong\u003e86\u003c/strong\u003e, 105291 (2023).\u003c/li\u003e\n\u003cli\u003eTsai, J.-Y. \u003cem\u003eet al.\u003c/em\u003e Lumbar disc herniation automatic detection in magnetic resonance imaging based on deep learning. \u003cem\u003eFrontiers in Bioengineering and Biotechnology\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 708137 (2021).\u003c/li\u003e\n\u003cli\u003eMa, S., Huang, Y., Che, X. \u0026amp; Gu, R. Faster RCNN-based detection of cervical spinal cord injury and disc degeneration. \u003cem\u003eJournal of applied clinical medical physics\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 235\u0026ndash;243 (2020).\u003c/li\u003e\n\u003cli\u003ePrisilla, A. A. \u003cem\u003eet al.\u003c/em\u003e An approach to the diagnosis of lumbar disc herniation using deep learning models. \u003cem\u003eFrontiers in Bioengineering and Biotechnology\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 1247112 (2023).\u003c/li\u003e\n\u003cli\u003eKang, H., Noh, D., Lee, S.-K., Choi, S. \u0026amp; Lee, K. Deep learning-based reconstruction can improve canine thoracolumbar magnetic resonance image quality and reduce slice thickness. \u003cem\u003eVeterinary Radiology \u0026amp; Ultrasound\u003c/em\u003e \u003cstrong\u003e64\u003c/strong\u003e, 1063\u0026ndash;1070 (2023).\u003c/li\u003e\n\u003cli\u003eBiercher, A. \u003cem\u003eet al.\u003c/em\u003e Using deep learning to detect spinal cord diseases on thoracolumbar magnetic resonance images of dogs. \u003cem\u003eFrontiers in veterinary science\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 721167 (2021).\u003c/li\u003e\n\u003cli\u003eZou, Z., Chen, K., Shi, Z., Guo, Y. \u0026amp; Ye, J. Object Detection in 20 Years: A Survey. \u003cem\u003eProceedings of the IEEE\u003c/em\u003e \u003cstrong\u003e111\u003c/strong\u003e, 257\u0026ndash;276 (2023).\u003c/li\u003e\n\u003cli\u003eRen, S., He, K., Girshick, R. \u0026amp; Sun, J. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. \u003cem\u003eIEEE Transactions on Pattern Analysis and Machine Intelligence\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 1137\u0026ndash;1149 (2017).\u003c/li\u003e\n\u003cli\u003eRedmon, J., Divvala, S., Girshick, R. \u0026amp; Farhadi, A. You only look once: Unified, real-time object detection. in \u003cem\u003eProceedings of the IEEE conference on computer vision and pattern recognition\u003c/em\u003e 779\u0026ndash;788 (2016).\u003c/li\u003e\n\u003cli\u003eGe, Z., Liu, S., Wang, F., Li, Z. \u0026amp; Sun, J. Yolox: Exceeding yolo series in 2021. \u003cem\u003earXiv preprint arXiv:2107.08430\u003c/em\u003e (2021).\u003c/li\u003e\n\u003cli\u003eChen, Q. \u003cem\u003eet al.\u003c/em\u003e You only look one-level feature. in \u003cem\u003eProceedings of the IEEE/CVF conference on computer vision and pattern recognition\u003c/em\u003e 13039\u0026ndash;13048 (2021).\u003c/li\u003e\n\u003cli\u003eTian, Y., Pu, L. Z. C. T., Singh, R., Burt, A. D. \u0026amp; Carneiro, G. One-Stage Five-Class Polyp Detection and Classification. in \u003cem\u003e2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)\u003c/em\u003e 70\u0026ndash;73 (2019). doi:10.1109/ISBI.2019.8759521.\u003c/li\u003e\n\u003cli\u003eCarranza-Garc\u0026iacute;a, M., Torres-Mateo, J., Lara-Ben\u0026iacute;tez, P. \u0026amp; Garc\u0026iacute;a-Guti\u0026eacute;rrez, J. On the Performance of One-Stage and Two-Stage Object Detectors in Autonomous Vehicles Using Camera Data. \u003cem\u003eRemote Sensing\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 89 (2021).\u003c/li\u003e\n\u003cli\u003eXu, J., Ren, H., Cai, S. \u0026amp; Zhang, X. An improved faster R-CNN algorithm for assisted detection of lung nodules. \u003cem\u003eComputers in Biology and Medicine\u003c/em\u003e \u003cstrong\u003e153\u003c/strong\u003e, 106470 (2023).\u003c/li\u003e\n\u003cli\u003eZhang, H., Chang, H., Ma, B., Wang, N. \u0026amp; Chen, X. Dynamic R-CNN: Towards high quality object detection via dynamic training. in \u003cem\u003eComputer Vision\u0026ndash;ECCV 2020: 16th European Conference, Glasgow, UK, August 23\u0026ndash;28, 2020, Proceedings, Part XV 16\u003c/em\u003e 260\u0026ndash;275 (Springer, 2020).\u003c/li\u003e\n\u003cli\u003eRedmon, J. \u0026amp; Farhadi, A. Yolov3: An incremental improvement. \u003cem\u003earXiv preprint arXiv:1804.02767\u003c/em\u003e (2018).\u003c/li\u003e\n\u003cli\u003eTian, Z., Shen, C., Chen, H. \u0026amp; He, T. Fcos: Fully convolutional one-stage object detection. in \u003cem\u003eProceedings of the IEEE/CVF international conference on computer vision\u003c/em\u003e 9627\u0026ndash;9636 (2019).\u003c/li\u003e\n\u003cli\u003eCai, Z. \u0026amp; Vasconcelos, N. Cascade R-CNN: High Quality Object Detection and Instance Segmentation. \u003cem\u003eIEEE Transactions on Pattern Analysis and Machine Intelligence\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 1483\u0026ndash;1498 (2021).\u003c/li\u003e\n\u003cli\u003eLu, X., Li, B., Yue, Y., Li, Q. \u0026amp; Yan, J. Grid r-cnn. in \u003cem\u003eProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition\u003c/em\u003e 7363\u0026ndash;7372 (2019).\u003c/li\u003e\n\u003cli\u003eVu, T., Jang, H., Pham, T. X. \u0026amp; Yoo, C. D. Cascade RPN: delving into high-quality region proposal network with adaptive convolution. in \u003cem\u003eProceedings of the 33rd International Conference on Neural Information Processing Systems\u003c/em\u003e (Curran Associates Inc., Red Hook, NY, USA, 2019).\u003c/li\u003e\n\u003cli\u003eChen, K. \u003cem\u003eet al.\u003c/em\u003e MMDetection: Open MMLab Detection Toolbox and Benchmark. \u003cem\u003earXiv preprint arXiv:1906.07155\u003c/em\u003e (2019).\u003c/li\u003e\n\u003cli\u003eHe, K., Zhang, X., Ren, S. \u0026amp; Sun, J. Deep residual learning for image recognition. in \u003cem\u003eProceedings of the IEEE conference on computer vision and pattern recognition\u003c/em\u003e 770\u0026ndash;778 (2016).\u003c/li\u003e\n\u003cli\u003eDeng, J. \u003cem\u003eet al.\u003c/em\u003e ImageNet: A large-scale hierarchical image database. in \u003cem\u003e2009 IEEE conference on computer vision and pattern recognition\u003c/em\u003e 248\u0026ndash;255 (IEEE, 2009).\u003c/li\u003e\n\u003cli\u003eRedmon, J. \u0026amp; Farhadi, A. YOLO9000: better, faster, stronger. in \u003cem\u003eProceedings of the IEEE conference on computer vision and pattern recognition\u003c/em\u003e 7263\u0026ndash;7271 (2017).\u003c/li\u003e\n\u003cli\u003eBochkovskiy, A., Wang, C.-Y. \u0026amp; Liao, H.-Y. M. YOLOv4: Optimal Speed and Accuracy of Object Detection. Preprint at https://doi.org/10.48550/arXiv.2004.10934 (2020).\u003c/li\u003e\n\u003cli\u003ePadilla, R., Netto, S. L. \u0026amp; da Silva, E. A. B. A Survey on Performance Metrics for Object-Detection Algorithms. in \u003cem\u003e2020 International Conference on Systems, Signals and Image Processing (IWSSIP)\u003c/em\u003e 237\u0026ndash;242 (2020). doi:10.1109/IWSSIP48289.2020.9145130.\u003c/li\u003e\n\u003cli\u003ePadilla, R., Passos, W. L., Dias, T. L. B., Netto, S. L. \u0026amp; da Silva, E. A. B. A Comparative Analysis of Object Detection Metrics with a Companion Open-Source Toolkit. \u003cem\u003eElectronics\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 279 (2021).\u003c/li\u003e\n\u003cli\u003eFarrell, M. \u0026amp; Fitzpatrick, N. Feline Intervertebral Disc Disease. in \u003cem\u003eAdvances in Intervertebral Disc Disease in Dogs and Cats\u003c/em\u003e 36\u0026ndash;49 (John Wiley \u0026amp; Sons, Ltd, 2015). doi:10.1002/9781118940372.ch6.\u003c/li\u003e\n\u003cli\u003eAmey, J. A., Liatis, T., Cherubini, G. B., De Decker, S. \u0026amp; Foreman, M. H. Outcomes of surgically and conservatively managed thoracolumbar and lumbosacral intervertebral disc herniations in cats. \u003cem\u003eJournal of Veterinary Internal Medicine\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 247\u0026ndash;257 (2024).\u003c/li\u003e\n\u003cli\u003eLin, T.-Y. \u003cem\u003eet al.\u003c/em\u003e Microsoft COCO: Common Objects in Context. in \u003cem\u003eComputer Vision \u0026ndash; ECCV 2014\u003c/em\u003e (eds. Fleet, D., Pajdla, T., Schiele, B. \u0026amp; Tuytelaars, T.) 740\u0026ndash;755 (Springer International Publishing, Cham, 2014). doi:10.1007/978-3-319-10602-1_48.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-3911180/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3911180/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntervertebral Disc Herniation (IVDH) is a common spinal disease in dogs, significantly impacting their health, mobility, and overall well-being. This study initiates an effort to automate the detection and localization of IVDH lesions in veterinary MRI scans, utilizing advanced artificial intelligence (AI) methods. A comprehensive canine IVDH dataset, comprising T2-weighted sagittal MRI images from 213 pet dogs of various breeds, ages, and sizes, was compiled and utilized to train and test the IVDH detection models. The experimental results showed that traditional two-stage detection models reliably outperformed one-stage models, including the recent You Only Look Once X (YOLOX) detector. In terms of methodology, this study introduced a novel spinal localization module, successfully integrated into different object detection models to enhance IVDH detection, achieving an average precision (AP) of up to 75.32%. Additionally, transfer learning was explored to adapt the IVDH detection model for a smaller feline dataset. Overall, this study provides insights into advancing AI for veterinary care, identifying challenges and exploring potential strategies for future development in veterinary radiology.\u003c/p\u003e","manuscriptTitle":"Exploring Deep Learning Strategies for Intervertebral Disc Herniation Detection on Veterinary MRI","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-02-22 19:39:13","doi":"10.21203/rs.3.rs-3911180/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-05-07T06:35:08+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-05-06T09:08:37+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-04-24T19:01:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"5423b8fb-1cb8-48b5-9445-1dcbef3b837e","date":"2024-04-15T07:02:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"e52f2d63-9e59-4f98-91aa-c0d88edcd613","date":"2024-04-03T01:59:44+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-04-01T14:44:23+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-03-22T16:38:51+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-02-20T07:15:42+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-02-20T07:03:12+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-01-30T16:09:33+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"714f3023-d2e6-4b6c-bb98-1fb6bef4297f","owner":[],"postedDate":"February 22nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":28856391,"name":"Physical sciences/Engineering/Biomedical engineering"},{"id":28856392,"name":"Health sciences/Health care/Medical imaging/Magnetic resonance imaging"},{"id":28856393,"name":"Health sciences/Neurology/Neurological disorders/Spinal cord diseases"},{"id":28856394,"name":"Biological sciences/Zoology"}],"tags":[],"updatedAt":"2024-07-20T00:31:33+00:00","versionOfRecord":{"articleIdentity":"rs-3911180","link":"https://doi.org/10.1038/s41598-024-67749-5","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-07-19 00:31:33","publishedOnDateReadable":"July 19th, 2024"},"versionCreatedAt":"2024-02-22 19:39:13","video":"","vorDoi":"10.1038/s41598-024-67749-5","vorDoiUrl":"https://doi.org/10.1038/s41598-024-67749-5","workflowStages":[]},"version":"v1","identity":"rs-3911180","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3911180","identity":"rs-3911180","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}