Dataset Information

Prediction of epidermal growth factor receptor (EGFR) mutation status in lung adenocarcinoma patients on computed tomography (CT) images using 3-dimensional (3D) convolutional neural network.

ABSTRACT:

Background

Noninvasively detecting epidermal growth factor receptor (EGFR) mutation status in lung adenocarcinoma patients before targeted therapy remains a challenge. This study aimed to develop a 3-dimensional (3D) convolutional neural network (CNN)-based deep learning model to predict EGFR mutation status using computed tomography (CT) images.

Methods

We retrospectively collected 660 patients from 2 large medical centers. The patients were divided into training (n=528) and external test (n=132) sets according to hospital source. The CNN model was trained in a supervised end-to-end manner, and its performance was evaluated using an external test set. To compare the performance of the CNN model, we constructed 1 clinical and 3 radiomics models. Furthermore, we constructed a comprehensive model combining the highest-performing radiomics and CNN models. The receiver operating characteristic (ROC) curves were used as primary measures of performance for each model. Delong test was used to compare performance differences between different models.

Results

Compared with the clinical [training set, area under the curve (AUC) =69.6%, 95% confidence interval (CI), 0.661-0.732; test set, AUC =68.4%, 95% CI, 0.609-0.752] and the highest-performing radiomics models (training set, AUC =84.3%, 95% CI, 0.812-0.873; test set, AUC =72.4%, 95% CI, 0.653-0.794) models, the CNN model (training set, AUC =94.3%, 95% CI, 0.920-0.961; test set, AUC =94.7%, 95% CI, 0.894-0.978) had significantly better predictive performance for predicting EGFR mutation status. In addition, compared with the comprehensive model (training set, AUC =95.7%, 95% CI, 0.942-0.971; test set, AUC =87.4%, 95% CI, 0.820-0.924), the CNN model had better stability.

Conclusions

The CNN model has excellent performance in non-invasively predicting EGFR mutation status in patients with lung adenocarcinoma and is expected to become an auxiliary tool for clinicians.

SUBMITTER: Zhang G

PROVIDER: S-EPMC11320524 | biostudies-literature | 2024 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Prediction of epidermal growth factor receptor (<i>EGFR</i>) mutation status in lung adenocarcinoma patients on computed tomography (CT) images using 3-dimensional (3D) convolutional neural network.

Zhang Guojin G Shang Lan L Cao Yuntai Y Zhang Jing J Li Shenglin S Qian Rong R Liu Huan H Zhang Zhuoli Z Pu Hong H Man Qiong Q Kong Weifang W

Quantitative imaging in medicine and surgery 20240730 8

<h4>Background</h4>Noninvasively detecting epidermal growth factor receptor (<i>EGFR</i>) mutation status in lung adenocarcinoma patients before targeted therapy remains a challenge. This study aimed to develop a 3-dimensional (3D) convolutional neural network (CNN)-based deep learning model to predict <i>EGFR</i> mutation status using computed tomography (CT) images.<h4>Methods</h4>We retrospectively collected 660 patients from 2 large medical centers. The patients were divided into training (n ...[more]

PMID: 39144003

Similar Datasets

Project description:PurposeWe developed an original, computer-aided diagnosis (CAD) software that subtracts the initial thoracic vertebral three-dimensional computed tomography (3D-CT) image from the follow-up 3D-CT image. The aim of this study was to investigate the efficacy of this CAD software during screening for vertebral metastases on follow-up CT images of primary lung cancer patients.Materials and methodsThe interpretation experiment included 30 sets of follow-up CT scans in primary lung cancer patients and was performed by two readers (readers A and B), who each had 2.5 years' experience reading CT images. In 395 vertebrae from C6 to L3, 46 vertebral metastases were identified as follows: osteolytic metastases (n = 17), osteoblastic metastases (n = 14), combined osteolytic and osteoblastic metastases (n = 6), and pathological fractures (n = 9). Thirty-six lesions were in the anterior component (vertebral body), and 10 lesions were in the posterior component (vertebral arch, transverse process, and spinous process). The area under the curve (AUC) by receiver operating characteristic (ROC) curve analysis and the sensitivity and specificity for detecting vertebral metastases were compared with and without CAD for each observer.ResultsReader A detected 47 abnormalities on CT images without CAD, and 33 of them were true-positive metastatic lesions. Using CAD, reader A detected 57 abnormalities, and 38 were true positives. The sensitivity increased from 0.717 to 0.826, and on ROC curve analysis, AUC with CAD was significantly higher than that without CAD (0.849 vs. 0.902, p = 0.021). Reader B detected 40 abnormalities on CT images without CAD, and 36 of them were true-positive metastatic lesions. Using CAD, reader B detected 44 abnormalities, and 39 were true positives. The sensitivity increased from 0.783 to 0.848, and AUC with CAD was nonsignificantly higher than that without CAD (0.889 vs. 0.910, p = 0.341). Both readers detected more osteolytic and osteoblastic metastases with CAD than without CAD.ConclusionOur temporal 3D-CT subtraction CAD software easily detected vertebral metastases on the follow-up CT images of lung cancer patients regardless of the osteolytic or osteoblastic nature of the lesions.

Project description:PurposeThe deformable nature of the liver can make focal treatment challenging and is not adequately addressed with simple rigid registration techniques. More advanced registration techniques can take deformations into account (eg, biomechanical modeling) but require segmentations of the whole liver for each scan, which is a time-intensive process. We hypothesize that fully convolutional networks can be used to rapidly and accurately autosegment the liver, removing the temporal bottleneck for biomechanical modeling.Methods and materialsManual liver segmentations on computed tomography scans from 183 patients treated at our institution and 30 scans from the Medical Image Computing & Computer Assisted Intervention challenges were collected for this study. Three architectures were investigated for rapid automated segmentation of the liver (VGG-16, DeepLabv3 +, and a 3-dimensional UNet). Fifty-six cases were set aside as a final test set for quantitative model evaluation. Accuracy of the autosegmentations was assessed using Dice similarity coefficient and mean surface distance. Qualitative evaluation was also performed by 3 radiation oncologists on 50 independent cases with previously clinically treated liver contours.ResultsThe mean (minimum-maximum) mean surface distance for the test groups with the final model, DeepLabv3 +, were as follows: μContrast(N = 17): 0.99 mm (0.47-2.2), μNon_Contrast(N = 19)l: 1.12 mm (0.41-2.87), and μMiccai(N = 30)t: 1.48 mm (0.82-3.96). The qualitative evaluation showed that 30 of 50 autosegmentations (60%) were preferred to manual contours (majority voting) in a blinded comparison, and 48 of 50 autosegmentations (96%) were deemed clinically acceptable by at least 1 reviewing physician.ConclusionsThe autosegmentations were preferred compared with manually defined contours in the majority of cases. The ability to rapidly segment the liver with high accuracy achieved in this investigation has the potential to enable the efficient integration of biomechanical model-based registration into a clinical workflow.

Project description:BackgroundDifferent pathological subtypes of lung adenocarcinoma lead to different treatment decisions and prognoses, and it is clinically important to distinguish invasive lung adenocarcinoma from preinvasive adenocarcinoma (adenocarcinoma in situ and minimally invasive adenocarcinoma). This study aims to investigate the performance of the deep learning approach based on high-resolution computed tomography (HRCT) images in the classification of tumor invasiveness and compare it with the performances of currently available approaches.MethodsIn this study, we used a deep learning approach based on 3D conventional networks to automatically predict the invasiveness of pulmonary nodules. A total of 901 early-stage non-small cell lung cancer patients who underwent surgical treatment at Shanghai Chest Hospital between November 2015 and March 2017 were retrospectively included and randomly assigned to a training set (n=814) or testing set 1 (n=87). We subsequently included 116 patients who underwent surgical treatment and intraoperative frozen section between April 2019 and January 2020 to form testing set 2. We compared the performance of our deep learning approach in predicting tumor invasiveness with that of intraoperative frozen section analysis and human experts (radiologists and surgeons).ResultsThe deep learning approach yielded an area under the receiver operating characteristic curve (AUC) of 0.946 for distinguishing preinvasive adenocarcinoma from invasive lung adenocarcinoma in the testing set 1, which is significantly higher than the AUCs of human experts (P<0.05). In testing set 2, the deep learning approach distinguished invasive adenocarcinoma from preinvasive adenocarcinoma with an AUC of 0.862, which is higher than that of frozen section analysis (0.755, P=0.043), senior thoracic surgeons (0.720, P=0.006), radiologists (0.766, P>0.05) and junior thoracic surgeons (0.768, P>0.05).ConclusionsWe developed a deep learning model that achieved comparable performance to intraoperative frozen section analysis in determining tumor invasiveness. The proposed method may contribute to clinical decisions related to the extent of surgical resection.

Project description:BackgroundEarly rectal neoplasms can be treated endoscopically with good prognosis, yet usually present with unspecific or an absence of signs and symptoms and are detected largely by invasive endoscopy with less compliance to screening. The purpose of this cross-sectional study was to explore the diagnostic value of dual-layer spectral detector computed tomography (DSCT) imaging for early rectal neoplasm.MethodsPatients who underwent DSCT for evaluation of rectal lesion or routine examination between September 2022 to September 2023 at West China Hospital were prospectively included and identified as group A (control, n=76), group B (rectal advanced adenomas and ≤T1 rectal cancer, n=59), and group C (≥T2 staging rectal cancer, n=74). Lesion visualization was graded to assess image quality. Spectral quantitative measurement, such as Hounsfield unit (HU)40 keV, HU70 keV, iodine concentration (IC), effective atomic number (Zeff), and the slope of spectral curve (λ), was analyzed and compared. Receiver operating characteristic (ROC) curves were generated to evaluate the diagnostic efficacy of spectral parameters. A comparison of ROC curves was applied to test the significance of differences between the area under the curves (AUCs).ResultsCompared to poly-energetic images (PEIs), the multiple parameters from DSCT were of greater capability to recognize rectal lesions. There were significant differences in HU40 keV (208.01±43.60 vs. 255.53±45.16), HU70 keV (87.06±18.55 vs. 100.78±18.26), IC [1.91 (1.71, 2.28) vs. 2.58±0.49], Zeff [8.33 (8.25, 8.50) vs. 8.61±0.20], and λ [3.80 (3.41, 4.52) vs. 5.16±1.00] between the early neoplastic lesions in rectum and the advanced rectal cancer (P<0.001). Significant correlations were found between the DSCT parameters and tumor staging (P<0.001). Furthermore, the AUCs of IC, Zeff, λ, and HUPEI were all above 0.90 for early rectal neoplasm detection, with additional capability of discriminating early rectal neoplasm from advanced rectal cancer.ConclusionsDSCT improved tumor conspicuity and the detection of the early rectal neoplastic lesion, suggesting that it is a promising screening tool in clinical practice.

Project description:BackgroundSkeletal class III malocclusion is among the most common dental and maxillofacial malformations. Three-dimensional (3D) printing technology has become widely applied in orthopaedics. The data source for 3D printing of maxillofacial bones is computed tomography (CT). The issue of the CT radiation dose caused by maxillofacial bone 3D printing has attracted increasing attention. This study aimed to explore the feasibility of low-dose CT technology in maxillofacial bone 3D printing and the clinical value of low-dose maxillofacial bone 3D printing.MethodsNinety patients with class III malocclusion who planned to undergo maxillofacial bone 3D printing and 3D-CT were prospectively enrolled and randomly divided into the conventional CT dose 3D printing group (Group A, n=28), low-CT dose 3D printing group (Group B, n=32) and 3D-CT control group (Group C, n=30). The quality of maxillofacial bone 3D printing was subjectively evaluated, and a Likert-scale questionnaire was used to assess the clinical value of maxillofacial bone 3D printing.ResultsNo significant differences in the general demographic characteristics were detected among Groups A, B, and C. Compared with that in Group A (0.8±0.1 mSv), the radiation effective dose (ED) in Group B (0.3±0.1 mSv) was reduced by approximately 63%. There were no significant differences between Groups A and B in 3D printing quality indices (including clarity, integrity, accuracy or artefacts) (all P>0.05). There were significantly higher subjective scores for the clinical value of maxillofacial bone 3D printing (Group A=4.1±0.5, 4.0±0.5, 4.0±0.4 and 4.1±0.5; Group B=4.0±0.5, 4.0±0.4, 4.0±0.5 and 4.0±0.5) than for 3D-CT (Group C=3.1±0.5, 3.1±0.4, 2.9±0.4 and 3.0±0.4) in diagnosing and classifying, formulating the surgical plan, simulating the surgical process, and predicting postoperative recovery (all P<0.05).ConclusionsLow-dose CT technology can be effectively applied for maxillofacial bone 3D printing, reducing the radiation dose without affecting the 3D printing quality. Maxillofacial bone 3D printing technology is superior to 3D-CT in class III malformations.