Quantitative Imaging in Medicine and Surgery最新文献

{"title":"An innovative bimodal computed tomography data-driven deep learning model for predicting aortic dissection: a multi-center study.","authors":"Ziqian Li, Lintao Chen, Shengxuming Zhang, Xiuming Zhang, Jing Zhang, Mingliang Ying, Jianyong Zhu, Ruiyang Li, Mingli Song, Zunlei Feng, Jianjun Zhang, Wenjie Liang","doi":"10.21037/qims-2024-2807","DOIUrl":"10.21037/qims-2024-2807","url":null,"abstract":"<p><strong>Background: </strong>Aortic dissection (AD) is a lethal emergency requiring prompt diagnosis. Current computed tomography angiography (CTA)-based diagnosis requires contrast agents, which expends time, whereas existing deep learning (DL) models only support single-modality inputs [non-contrast computed tomography (CT) or CTA]. In this study, we propose a bimodal DL framework to independently process both types, enabling dual-path detection and improving diagnostic efficiency.</p><p><strong>Methods: </strong>Patients who underwent non-contrast CT and CTA from February 2016 to September 2021 were retrospectively included from three institutions, including the First Affiliated Hospital, Zhejiang University School of Medicine (Center I), Zhejiang Hospital (Center II), and Yiwu Central Hospital (Center III). A two-stage DL model for predicting AD was developed. The first stage used an aorta detection network (AoDN) to localize the aorta in non-contrast CT or CTA images. Image patches that contained detected aorta were cut from CT images and combined to form an image patch sequence, which was inputted to an aortic dissection diagnosis network (ADDiN) to diagnose AD in the second stage. The following performances were assessed: aorta detection and diagnosis using average precision at the intersection over union threshold 0.5 (AP@0.5) and area under the receiver operating characteristic curve (AUC).</p><p><strong>Results: </strong>The first cohort, comprising 102 patients (53±15 years, 80 men) from two institutions, was used for the AoDN, whereas the second cohort, consisting of 861 cases (55±15 years, 623 men) from three institutions, was used for the ADDiN. For the AD task, the AoDN achieved AP@0.5 99.14% on the non-contrast CT test set and 99.34% on the CTA test set, respectively. For the AD diagnosis task, the ADDiN obtained an AUCs of 0.98 on the non-contrast CT test set and 0.99 on the CTA test set.</p><p><strong>Conclusions: </strong>The proposed bimodal CT data-driven DL model accurately diagnoses AD, facilitating prompt hospital diagnosis and treatment of AD.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8359-8371"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397666/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparing respiratory-triggered T2WI MRI with an artificial intelligence-assisted technique and motion-suppressed respiratory-triggered T2WI in abdominal imaging.","authors":"Nan Wang, Yuhui Liu, Jiangnan Ran, Qi An, Lihua Chen, Ying Zhao, Dan Yu, Ailian Liu, Lina Zhuang, Qingwei Song","doi":"10.21037/qims-2025-71","DOIUrl":"10.21037/qims-2025-71","url":null,"abstract":"<p><strong>Background: </strong>Magnetic resonance imaging (MRI) plays a crucial role in the diagnosis of abdominal conditions. A comprehensive assessment, especially of the liver, requires multi-planar T2-weighted sequences. To mitigate the effect of respiratory motion on image quality, the combination of acquisition and reconstruction with motion suppression (ARMS) and respiratory triggering (RT) is commonly employed. While this method maintains image quality, it does so at the expense of longer acquisition times. We evaluated the effectiveness of free-breathing, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging (ACS-RT T2WI) compared to conventional acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging (ARMS-RT T2WI) in abdominal MRI, assessing both qualitative and quantitative measures of image quality and lesion detection.</p><p><strong>Methods: </strong>In this retrospective study, 334 patients with upper abdominal discomfort were examined on a 3.0T MRI system. Each patient underwent both ARMS-RT T2WI and ACS-RT T2WI. Image quality was analyzed by two independent readers using a five-point Likert scale. The quantitative measurements included the signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), peak signal-to-noise ratio (PSNR), and sharpness. Lesion detection rates and contrast ratios (CRs) were also evaluated for liver, biliary system, and pancreatic lesions.</p><p><strong>Results: </strong>There ACS-RT T2WI protocol had a significantly reduced median scanning time compared to the ARMS-RT T2WI protocol (148.22±38.37 <i>vs.</i> 13.86±1.72 seconds). However, ARMS-RT T2WI had a higher PSNR than ACS-RT T2WI (39.87±2.72 <i>vs.</i> 38.69±3.00, P<0.05). Of the 201 liver lesions, ARMS-RT T2WI detected 193 (96.0%) and ACS-RT T2WI detected 192 (95.5%) (P=0.787). Of the 97 biliary system lesions, ARMS-RT T2WI detected 92 (94.8%) and ACS-RT T2WI detected 94 (96.9%) (P=0.721). Of the 110 pancreatic lesions, ARMS-RT T2WI detected 102 (92.7%) and ACS-RT T2WI detected 104 (94.5%) (P=0.784). The CR analysis showed the superior performance of ACS-RT T2WI in certain lesion types (hemangioma, 0.58±0.11 <i>vs.</i> 0.55±0.12; biliary tumor, 0.47±0.09 <i>vs.</i> 0.38±0.09; pancreatic cystic lesions, 0.59±0.12 <i>vs.</i> 0.48±0.14; pancreatic cancer, 0.48±0.18 <i>vs.</i> 0.43±0.17), but no significant difference was found in others like focal nodular hyperplasia (FNH), hepatapostema, hepatocellular carcinoma (HCC), cholangiocarcinoma, metastatic tumors, and biliary calculus.</p><p><strong>Conclusions: </strong>ACS-RT T2WI ensures clinical reliability with a substantial scan time reduction (>80%). Despite minor losses in detail and SNR reduction, ACS-RT T2WI does not impair lesion detection, marking its efficacy in abdominal imaging.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"7761-7773"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397657/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978582","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Association of decreased thalamic neurotransmitter level with sensorimotor tract damage in patients with relapsing-remitting multiple sclerosis.","authors":"Yan Xie, Yujie Ding, Shaolong Wu, Yan Zhang, Hongquan Zhu, Yuanhao Li, Xiaoxiao Zhang, Wenzhen Zhu","doi":"10.21037/qims-2025-219","DOIUrl":"10.21037/qims-2025-219","url":null,"abstract":"<p><strong>Background: </strong>Thalamic microstructural damage and neurotransmission dysfunction are present in patients with multiple sclerosis (MS). The aim of this study was to investigate the relationship of altered γ-aminobutyric acid (GABA) and glutamate + glutamine (Glx) levels in the thalamus with the white-matter (WM) microstructural damage of the sensorimotor tract in patients with relapsing-remitting MS (RRMS).</p><p><strong>Methods: </strong>In this cross-sectional study, 50 patients with RRMS and 43 healthy controls (HCs) were scanned using Mescher-Garwood point resolved spectroscopy (MEGA-PRESS) to quantify the GABA+ and Glx level of the thalamus. Metrics derived from diffusion tensor imaging (DTI) were calculated to reflect the degree of WM microstructural damage of the sensorimotor tract. The correlation between neurotransmitter level and diffusion metrics was determined in patients with RRMS and HCs, respectively.</p><p><strong>Results: </strong>Thalamic GABA+ and Glx levels were significantly decreased in patients with RRMS as compared with HCs (GABA+: 2.859±0.451 <i>vs.</i> 3.092±0.283 IU, P=0.002; Glx: 5.787±1.307 <i>vs.</i> 6.439±0.680 IU, P=0.002), and the neurotransmitter levels were significantly and negatively correlated with total lesion volume and disease duration in patients with RRMS (P<0.05). With the exception of the tract of right supplementary motor area, other sensorimotor tracts of patients with RRMS showed extensive WM microstructural damage. In addition, there was a significant correlation between decreased thalamic GABA+ and Glx levels and sensorimotor tract damage in patients with RRMS (corrected P<0.05). Analysis in HCs showed that the thalamic neurotransmitter level was not correlated with diffusion metrics in any of the sensorimotor tracts.</p><p><strong>Conclusions: </strong>Neurotransmitters may play an important role in the pathophysiologic mechanisms of MS. Our study suggests an association between altered GABA and glutamate levels in deep gray-matter and WM microstructural damage in the sensorimotor tract.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8040-8054"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397645/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Prenatal diagnosis of retrorenal left-sided inferior vena cava without hepatic segment using ultrasound and spatiotemporal image correlation: a rare case description.","authors":"Pujuan Jia, Yanzhao Wang, Minhong Zhang, Xuelu Feng, Xiaxia Cheng, Bin Ma","doi":"10.21037/qims-2024-2468","DOIUrl":"10.21037/qims-2024-2468","url":null,"abstract":"","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8742-8745"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397701/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978554","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Endoscopic endonasal transsphenoidal approach of pituitary macroadenoma and optic canal stenosis in a patient with McCune-Albright syndrome.","authors":"Leonardo A S Beck, Fabiano Reis, Heraldo Mendes Garmes, Thiago L Infanger Serrano, Fabio Lau, Marcelo Hamilton Sampaio, Mateus Dal Fabbro","doi":"10.21037/qims-2025-176","DOIUrl":"10.21037/qims-2025-176","url":null,"abstract":"","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8703-8708"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397750/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978572","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fractional flow reserve calculation optimized by myocardial computed tomography perfusion information.","authors":"Peiming Qin, Yan Yi, Cheng Xu, Limiao Zou, Fenggang Jia, Jian Guo, Ming Wang, Yan Zhang, Ziquan Wang, Pei Dong, Dijia Wu, Xiaodong Wang, Yining Wang","doi":"10.21037/qims-24-2172","DOIUrl":"10.21037/qims-24-2172","url":null,"abstract":"<p><strong>Background: </strong>Current computed tomography angiography-derived fractional flow reserve (CT-FFR) diagnosis leaves room for improvement in diagnosing coronary heart disease. In this study, the computed fluid dynamics boundary condition optimization method was used to calculate CT-FFR aiming to improve the diagnostic accuracy of CT-FFR for coronary heart disease.</p><p><strong>Methods: </strong>The two enhancement approaches are as follows: (A) inlet flow optimization, which involves determining the total coronary inlet flow rate by summing the myocardial blood flow (MBF) across the entire left ventricular myocardium; and (B) inlet & outlet flow optimization: building upon method A, where the outlet flow of coronary artery branches is calculated through blood supply area analysis.</p><p><strong>Results: </strong>A total of 100 fractional flow reserve pressure guide wire measurement sites from 47 cases were used to evaluate the above two methods comparing with the traditional computed fluid dynamics method without computed tomography perfusion (CTP) images. In traditional method, the accuracy was 88%, the sensitivity was 91.4% (95% confidence interval: 75.8-97.7%), and the specificity was 86.2% (95% confidence interval: 74.8-93.1%). In Method A, the accuracy improved by 5% (93%), the sensitivity remained unchanged (91.4%, 95% confidence interval: 75.8-97.7%), and the specificity increased by 7.6% (93.8%, 95% confidence interval: 84.2-98%). In Method B, the accuracy increased by 6% (94%), the sensitivity increased to 100% (95% confidence interval: 87.7-100%), and the specificity increased by 4.8% (94%, 95% confidence interval: 80.3-96.2%).</p><p><strong>Conclusions: </strong>The computed fluid dynamics calculation, guided by MBF values from stress CTP imaging, helps enhance the consistency between CT-FFR calculation and invasive fractional flow reserve measurements.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"7909-7921"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397671/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978573","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Rare case of recurrent refractory multiple myeloma with lipid deposition in the vertebral body: a case description.","authors":"Yanju Li, Ying Chen, Yang Liu, Yuanyuan Pei, Kaiji Zhang, Feiqing Wang","doi":"10.21037/qims-24-2207","DOIUrl":"10.21037/qims-24-2207","url":null,"abstract":"","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8737-8741"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397708/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978743","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improved image quality and diagnostic performance of coronary computed tomography angiography-derived fractional flow reserve with super-resolution deep learning reconstruction.","authors":"Li-Miao Zou, Cheng Xu, Min Xu, Ke-Ting Xu, Ming Wang, Yun Wang, Yi-Ning Wang","doi":"10.21037/qims-24-2075","DOIUrl":"10.21037/qims-24-2075","url":null,"abstract":"<p><strong>Background: </strong>Super-resolution deep learning reconstruction (SR-DLR) algorithm has emerged as a promising image reconstruction technique for improving the image quality of coronary computed tomography angiography (CCTA) and ensuring accurate CCTA-derived fractional flow reserve (CT-FFR) assessments even in problematic scenarios (e.g., the presence of heavily calcified plaque and stent implantation). Therefore, the purposes of this study were to evaluate the image quality of CCTA obtained with SR-DLR in comparison with conventional reconstruction methods and to investigate the diagnostic performances of different reconstruction approaches based on CT-FFR.</p><p><strong>Methods: </strong>Fifty patients who underwent CCTA and subsequent invasive coronary angiography (ICA) were retrospectively included. All images were reconstructed with hybrid iterative reconstruction (HIR), model-based iterative reconstruction (MBIR), conventional deep learning reconstruction (C-DLR), and SR-DLR algorithms. Objective parameters and subjective scores were compared. Among the patients, 22-comprising 45 lesions-had invasive FFR results as a reference, and the diagnostic performance of different reconstruction approaches based on CT-FFR were compared.</p><p><strong>Results: </strong>SR-DLR achieved the lowest image noise, highest signal-to-noise ratio (SNR), and best edge sharpness (all P values <0.05), as well as the best subjective scores from both reviewers (all P values <0.001). With FFR serving as a reference, the specificity and positive predictive value (PPV) were improved as compared with HIR and C-DLR (72% <i>vs.</i> 36-44% and 73% <i>vs.</i> 53-58%, respectively); moreover, SR-DLR improved the sensitivity and negative predictive value (NPV) as compared to MBIR (95% <i>vs.</i> 70% and 95% <i>vs.</i> 68%, respectively; all P values <0.05). The overall diagnostic accuracy and area under the curve (AUC) for SR-DLR were significantly higher than those of the HIR, MBIR, and C-DLR algorithms (82% <i>vs.</i> 60-67% and 0.84 <i>vs.</i> 0.61-0.70, respectively; all P values <0.05).</p><p><strong>Conclusions: </strong>SR-DLR had the best image quality for both objective and subjective evaluation. The diagnostic performances of CT-FFR were improved by SR-DLR, enabling more accurate assessment of flow-limiting lesions.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8541-8552"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397712/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978745","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel RANSAC-based method for detecting and estimating externally attached marker spheres in craniomaxillofacial CT images.","authors":"Yonghui Li, Han Zhang, Weili Shi, Wei He, Yu Miao, Guodong Wei, Zhengang Jiang","doi":"10.21037/qims-2025-386","DOIUrl":"10.21037/qims-2025-386","url":null,"abstract":"<p><strong>Background: </strong>The precision of image-physical space registration using spherical markers in craniomaxillofacial surgical navigation significantly depends on the accurate estimation of spherical parameters from computed tomography (CT) images. However, this estimation is susceptible to the abnormal points caused by artifacts, instruments interference, and other factors. To address these challenges, this study proposes a robust method to improve reproducibility in results and achieve higher accuracy on low inlier ratio data, thereby meeting the requirements of high-precision surgical applications.</p><p><strong>Methods: </strong>Firstly, potential marker sphere regions are isolated from CT images. Next, we propose the Local Evaluation and Optimization RANdom SAmple Consensus (LEO-RANSAC) algorithm to refine the detection of the spherical parameters. This technique introduces a metric that combines multi-level adaptive curvature and local solution to filter local models, and adopts an equidistance adjustment mechanism to improve the accuracy of the so-far-the-best model. Lastly, a custom-designed equipment is utilized to measure the fiducial localization error (FLE), and a skull phantom study is utilized to evaluate the fiducial registration error (FRE) and the target registration error (TRE).</p><p><strong>Results: </strong>The proposed method was evaluated on 72-point clouds with inlier ratio ranging from 30% to 90%. After repeating 100 independent experiments, the deviations of the maximum of FLEs for six different configurations were 0.40±0.25, 0.52±0.35, 0.58±0.35, 0.53±0.25, 0.51±0.28, and 0.39±0.31 mm, respectively. Analysis of 72 results showed that 87.50% of the maximum of FLEs were less than 0.9 mm, and 95.83% of the variances of FLEs were less than 0.01. In a skull phantom study involving 3 different datasets, the FREs were 0.4222, 0.5223, and 0.372 mm, respectively, whereas the TREs were 0.8546, 0.9471, and 0.8537 mm during real-time guidance, respectively.</p><p><strong>Conclusions: </strong>The results demonstrate that our method outperforms existing approaches in terms of accuracy and reliability, highlighting its potential applicability in craniomaxillofacial surgical navigation.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"8023-8039"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397674/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978678","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Influence of preprocessing of stimulated Raman scattering images on the performance of deep neural networks for detecting cancer tissue.","authors":"Andreas Weber, Kathrin Enderle-Ammour, Karl-Moritz Schröder, Marc C Metzger, Leonard Simon Brandenburg, Jürgen Beck, Jakob Straehle, David Steybe, Mohamed Hassan, Severin Schmid, Birte Ohm, Martin Werner, Bernward Passlick, Rainer Schmelzeisen, Uyen-Thao Le, Peter Bronsert","doi":"10.21037/qims-2024-2608","DOIUrl":"10.21037/qims-2024-2608","url":null,"abstract":"<p><strong>Background: </strong>Images obtained from stimulated Raman scattering can be used to identify histomorphologically relevant information intraoperatively. In order to leverage deep learning algorithms for distinguishing tumoral and non-tumoral tissue, data preprocessing remains a crucial task and may affect the classification performance. To date, the effect of different preprocessing techniques on deep learning algorithm performance is unclear. This study aims to make a contribution to closing this knowledge gap.</p><p><strong>Methods: </strong>To investigate the influence of different preprocessing techniques of images obtained from stimulated Raman scattering, six deep learning architectures (VGG19, ResNet50, InceptionResNetV2, Xception, ConvNeXt and Vision Transformer) and five different preprocessing procedures were compared. For this, annotated datasets comprising 542 images of tissue samples obtained from patients with oral squamous cell carcinoma and non-small cell lung carcinoma were used for network training. Each network was trained five times for 40 epochs. Performance metrics balanced accuracy, precision, recall and F1-score were recorded. Class activation and attention maps were used to highlight on which input pixels a prediction is based.</p><p><strong>Results: </strong>A scaling of the original pixel values of stimulated Raman scattering images to the range [0, 1] yielded a higher and more stable overall classification performance across the neural networks when compared to more sophisticated and computationally expensive methods [ <math> <mrow> <mover><mrow><mi>F</mi> <mn>1</mn></mrow> <mo>¯</mo></mover> </mrow> </math> =0.8327; standard deviation (SD) =0.0622 on scaled dataset and <math> <mrow> <mover><mrow><mi>F</mi> <mn>1</mn></mrow> <mo>¯</mo></mover> </mrow> </math> =0.7213 (SD =0.2315) on complex preprocessed dataset; P≤0.05]. Absolute performance was best on stimulated Raman histology images ( <math> <mrow> <mover><mrow><mi>F</mi> <mn>1</mn></mrow> <mo>¯</mo></mover> </mrow> </math> =0.8478; SD =0.1487).</p><p><strong>Conclusions: </strong>This study shows that preprocessing of pixel values of stimulated Raman scattering images can have a great impact on the performance and the stability of deep learning algorithms when applied for classification of cancer tissue.</p>","PeriodicalId":54267,"journal":{"name":"Quantitative Imaging in Medicine and Surgery","volume":"15 9","pages":"7711-7726"},"PeriodicalIF":2.3,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12397652/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144978694","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}