2023 Survey on User Experience of Artificial Intelligence Software in Radiology by the Korean Society of Radiology.

OBJECTIVE: In Korea, radiology has been positioned towards the early adoption of artificial intelligence-based software as medical devices (AI-SaMDs); however, little is known about the current usage, implementation, and future needs of AI-SaMDs. We surveyed the current trends and expectations for AI-SaMDs among members of the Korean Society of Radiology (KSR). MATERIALS AND METHODS: An anonymous and voluntary online survey was open to all KSR members between April 17 and May 15, 2023. The survey was focused on the experiences of using AI-SaMDs, patterns of usage, levels of satisfaction, and expectations regarding the use of AI-SaMDs, including the roles of the industry, government, and KSR regarding the clinical use of AI-SaMDs. RESULTS: Among the 370 respondents (response rate: 7.7% [370/4792]; 340 board-certified radiologists; 210 from academic institutions), 60.3% (223/370) had experience using AI-SaMDs. The two most common use-case of AI-SaMDs among the respondents were lesion detection (82.1%, 183/223), lesion diagnosis/classification (55.2%, 123/223), with the target imaging modalities being plain radiography (62.3%, 139/223), CT (42.6%, 95/223), mammography (29.1%, 65/223), and MRI (28.7%, 64/223). Most users were satisfied with AI-SaMDs (67.6% [115/170, for improvement of patient management] to 85.1% [189/222, for performance]). Regarding the expansion of clinical applications, most respondents expressed a preference for AI-SaMDs to assist in detection/diagnosis (77.0%, 285/370) and to perform automated measurement/quantification (63.5%, 235/370). Most respondents indicated that future development of AI-SaMDs should focus on improving practice efficiency (81.9%, 303/370) and quality (71.4%, 264/370). Overall, 91.9% of the respondents (340/370) agreed that there is a need for education or guidelines driven by the KSR regarding the use of AI-SaMDs. CONCLUSION: The penetration rate of AI-SaMDs in clinical practice and the corresponding satisfaction levels were high among members of the KSR. Most AI-SaMDs have been used for lesion detection, diagnosis, and classification. Most respondents requested KSR-driven education or guidelines on the use of AI-SaMDs.

Computed Tomography Radiomics for Preoperative Prediction of Spread Through Air Spaces in the Early Stage of Surgically Resected Lung Adenocarcinomas.

PURPOSE: To assess the added value of radiomics models from preoperative chest CT in predicting the presence of spread through air spaces (STAS) in the early stage of surgically resected lung adenocarcinomas using multiple validation datasets. MATERIALS AND METHODS: This retrospective study included 550 early-stage surgically resected lung adenocarcinomas in 521 patients, classified into training, test, internal validation, and temporal validation sets (n=211, 90, 91, and 158, respectively). Radiomics features were extracted from the segmented tumors on preoperative chest CT, and a radiomics score (Rad-score) was calculated to predict the presence of STAS. Diagnostic performance of the conventional model and the combined model, based on a combination of conventional and radiomics features, for the diagnosis of the presence of STAS were compared using the area under the curve (AUC) of the receiver operating characteristic curve. RESULTS: Rad-score was significantly higher in the STAS-positive group compared to the STAS-negative group in the training, test, internal, and temporal validation sets. The performance of the combined model was significantly higher than that of the conventional model in the training set {AUC: 0.784 [95% confidence interval (CI): 0.722-0.846] vs. AUC: 0.815 (95% CI: 0.759-0.872), p=0.042}. In the temporal validation set, the combined model showed a significantly higher AUC than that of the conventional model (p=0.001). The combined model showed a higher AUC than the conventional model in the test and internal validation sets, albeit with no statistical significance. CONCLUSION: A quantitative CT radiomics model can assist in the non-invasive prediction of the presence of STAS in the early stage of lung adenocarcinomas.

Influence of computed tomography slice thickness on deep learning-based, automatic coronary artery calcium scoring software performance.

Background: The influence of computed tomography (CT) slice thickness on the accuracy of deep learning (DL)-based, automatic coronary artery calcium (CAC) scoring software has not been explored yet. Methods: This retrospective study included 844 subjects (477 men, mean age of 58.9±10.7 years) who underwent electrocardiogram (ECG)-gated CAC scoring CT scans with 1.5 and 3 mm slice thickness values between September 2013 and October 2020. Automatic CAC scoring was performed using DL-based software (3D patch-based U-Net architectures). Manual CAC scoring was set as the reference standard. The reliability of automatic CAC scoring was evaluated using intraclass correlation coefficients (ICCs) for both the 1.5 and 3 mm datasets. The agreement of CAC severity categories [Agatston score (AS) 0, 1-100, 101-400, >400] between automatic CAC scoring and the reference standard was analyzed using weighted kappa (κ) statistics for both 1.5 and 3 mm datasets. Results: The CAC scoring agreement between the automatic CAC scoring and reference standard was excellent (ICC 0.982 for 1.5 mm, 0.969 for 3 mm, respectively). The categorical agreement of CAC severity between two methods was excellent for both 1.5 and 3 mm scans, with better agreement for 3 mm scans (weighted κ: 0.851 and 0.961, 95% confidence intervals: 0.823-0.879 and 0.945-0.974, respectively). Conclusions: Automatic CAC scoring shows excellent agreement with the reference standard for both 1.5 and 3 mm scans but results in lower agreement in the CAC severity category for 1.5 mm scans.

A machine learning model for predicting hepatocellular carcinoma risk in patients with chronic hepatitis B.

BACKGROUND: Machine learning (ML) algorithms can be used to overcome the prognostic performance limitations of conventional hepatocellular carcinoma (HCC) risk models. We established and validated an ML-based HCC predictive model optimized for patients with chronic hepatitis B (CHB) infections receiving antiviral therapy (AVT). METHODS: Treatment-naïve CHB patients who were started entecavir (ETV) or tenofovir disoproxil fumarate (TDF) were enrolled. We used a training cohort (n = 960) to develop a novel ML model that predicted HCC development within 5 years and validated the model using an independent external cohort (n = 1937). ML algorithms consider all potential interactions and do not use predefined hypotheses. RESULTS: The mean age of the patients in the training cohort was 48 years, and most patients (68.9%) were men. During the median 59.3 (interquartile range 45.8-72.3) months of follow-up, 69 (7.2%) patients developed HCC. Our ML-based HCC risk prediction model had an area under the receiver-operating characteristic curve (AUC) of 0.900, which was better than the AUCs of CAMD (0.778) and REAL B (0.772) (both p < .05). The better performance of our model was maintained (AUC = 0.872 vs. 0.788 for CAMD and 0.801 for REAL B) in the validation cohort. Using cut-off probabilities of 0.3 and 0.5, the cumulative incidence of HCC development differed significantly among the three risk groups (p < .001). CONCLUSIONS: Our new ML model performed better than models in terms of predicting the risk of HCC development in CHB patients receiving AVT.

Fully automated radiomics-based machine learning models for multiclass classification of single brain tumors: Glioblastoma, lymphoma, and metastasis.

BACKGROUND AND PURPOSE: To investigate the diagnostic performance of fully automated radiomics-based models for multiclass classification of a single enhancing brain tumor among glioblastoma, central nervous system lymphoma, and metastasis. MATERIALS AND METHODS: The training and test sets were comprised of 538 cases (300 glioblastomas, 73 lymphomas, and 165 metastases) and 169 cases (101 glioblastomas, 29 lymphomas, and 39 metastases), respectively. After fully automated segmentation, radiomic features were extracted. Three conventional machine learning classifiers, including least absolute shrinkage and selection operator (LASSO), adaptive boosting (Adaboost), and support vector machine with the linear kernel (SVC), combined with one of four feature selection methods, including forward sequential feature selection, F score, mutual information, and LASSO, were trained. Additionally, one ensemble classifier based on the three classifiers was used. The diagnostic performance of the optimized models was tested in the test set using the accuracy, F1-macro score, and the area under the receiver operating characteristic curve (AUCROC). RESULTS: The best performance was achieved when the LASSO was used as a feature selection method. In the test set, the best performance was achieved by the ensemble classifier, showing an accuracy of 76.3% (95% CI, 70.0-82.7), a F1-macro score of 0.704, and an AUCROC of 0.878. CONCLUSION: Our fully automated radiomics-based models for multiclass classification might be useful for differential diagnosis of a single enhancing brain tumor with a good diagnostic performance and generalizability.

Clinical application of deep learning-based synthetic CT from real MRI to improve dose planning accuracy in Gamma Knife radiosurgery: a proof of concept study.

Dose planning for Gamma Knife radiosurgery (GKRS) uses the magnetic resonance (MR)-based tissue maximum ratio (TMR) algorithm, which calculates radiation dose without considering heterogeneous radiation attenuation in the tissue. In order to plan the dose considering the radiation attenuation, the Convolution algorithm should be used, and additional radiation exposure for computed tomography (CT) and registration errors between MR and CT are entailed. This study investigated the clinical feasibility of synthetic CT (sCT) from GKRS planning MR using deep learning. The model was trained using frame-based contrast-enhanced T1-weighted MR images and corresponding CT slices from 54 training subjects acquired for GKRS planning. The model was applied prospectively to 60 lesions in 43 patients including benign tumor such as meningioma and pituitary adenoma, metastatic brain tumors, and vascular disease of various location for evaluating the model and its application. We evaluated the sCT and compared between treatment plans made with MR only (TMR 10 plan), MR and real CT (rCT; Convolution with rCT [Conv-rCT] plan), and MR and synthetic CT (Convolution with sCT [Conv-sCT] plan). The mean absolute error (MAE) of 43 sCT was 107.35 ± 16.47 Hounsfield units. The TMR 10 treatment plan differed significantly from plans made by Conv-sCT and Conv-rCT. However, the Conv-sCT and Conv-rCT plans were similar. This study showed the practical applicability of deep learning based on sCT in GKRS. Our results support the possibility of formulating GKRS treatment plans while considering radiation attenuation in the tissue using GKRS planning MR and no radiation exposure.

Radiomics machine learning study with a small sample size: Single random training-test set split may lead to unreliable results.

This study aims to determine how randomly splitting a dataset into training and test sets affects the estimated performance of a machine learning model and its gap from the test performance under different conditions, using real-world brain tumor radiomics data. We conducted two classification tasks of different difficulty levels with magnetic resonance imaging (MRI) radiomics features: (1) "Simple" task, glioblastomas [n = 109] vs. brain metastasis [n = 58] and (2) "difficult" task, low- [n = 163] vs. high-grade [n = 95] meningiomas. Additionally, two undersampled datasets were created by randomly sampling 50% from these datasets. We performed random training-test set splitting for each dataset repeatedly to create 1,000 different training-test set pairs. For each dataset pair, the least absolute shrinkage and selection operator model was trained and evaluated using various validation methods in the training set, and tested in the test set, using the area under the curve (AUC) as an evaluation metric. The AUCs in training and testing varied among different training-test set pairs, especially with the undersampled datasets and the difficult task. The mean (±standard deviation) AUC difference between training and testing was 0.039 (±0.032) for the simple task without undersampling and 0.092 (±0.071) for the difficult task with undersampling. In a training-test set pair with the difficult task without undersampling, for example, the AUC was high in training but much lower in testing (0.882 and 0.667, respectively); in another dataset pair with the same task, however, the AUC was low in training but much higher in testing (0.709 and 0.911, respectively). When the AUC discrepancy between training and test, or generalization gap, was large, none of the validation methods helped sufficiently reduce the generalization gap. Our results suggest that machine learning after a single random training-test set split may lead to unreliable results in radiomics studies especially with small sample sizes.

Radiomics With Ensemble Machine Learning Predicts Dopamine Agonist Response in Patients With Prolactinoma.

CONTEXT: Early identification of the response of prolactinoma patients to dopamine agonists (DA) is crucial in treatment planning. OBJECTIVE: To develop a radiomics model using an ensemble machine learning classifier with conventional magnetic resonance images (MRIs) to predict the DA response in prolactinoma patients. DESIGN: Retrospective study. SETTING: Severance Hospital, Seoul, Korea. PATIENTS: A total of 177 prolactinoma patients who underwent baseline MRI (109 DA responders and 68 DA nonresponders) were allocated to the training (n = 141) and test (n = 36) sets. Radiomic features (n = 107) were extracted from coronal T2-weighed MRIs. After feature selection, single models (random forest, light gradient boosting machine, extra-trees, quadratic discrimination analysis, and linear discrimination analysis) with oversampling methods were trained to predict the DA response. A soft voting ensemble classifier was used to achieve the final performance. The performance of the classifier was validated in the test set. RESULTS: The ensemble classifier showed an area under the curve (AUC) of 0.81 [95% confidence interval (CI), 0.74-0.87] in the training set. In the test set, the ensemble classifier showed an AUC, accuracy, sensitivity, and specificity of 0.81 (95% CI, 0.67-0.96), 77.8%, 78.6%, and 77.3%, respectively. The ensemble classifier achieved the highest performance among all the individual models in the test set. CONCLUSIONS: Radiomic features may be useful biomarkers to predict the DA response in prolactinoma patients.

Using deep learning to predict temporomandibular joint disc perforation based on magnetic resonance imaging.

The goal of this study was to develop a deep learning-based algorithm to predict temporomandibular joint (TMJ) disc perforation based on the findings of magnetic resonance imaging (MRI) and to validate its performance through comparison with previously reported results. The study objects were obtained by reviewing medical records from January 2005 to June 2018. 299 joints from 289 patients were divided into perforated and non-perforated groups based on the existence of disc perforation confirmed during surgery. Experienced observers interpreted the TMJ MRI images to extract features. Data containing those features were applied to build and validate prediction models using random forest and multilayer perceptron (MLP) techniques, the latter using the Keras framework, a recent deep learning architecture. The area under the receiver operating characteristic (ROC) curve (AUC) was used to compare the performances of the models. MLP produced the best performance (AUC 0.940), followed by random forest (AUC 0.918) and disc shape alone (AUC 0.791). The MLP and random forest were also superior to previously reported results using MRI (AUC 0.808) and MRI-based nomogram (AUC 0.889). Implementing deep learning showed superior performance in predicting disc perforation in TMJ compared to conventional methods and previous reports.

Development of a deep learning-based algorithm for the automatic detection and quantification of aortic valve calcium.

PURPOSE: We aimed to develop a deep learning (DL)-based algorithm for automated quantification of aortic valve calcium (AVC) from non-enhanced electrocardiogram-gated cardiac CT scans and compare performance of DL-measured AVC volume and Agatston score with those of visual gradings by radiologist readers for classification of AVC severity. METHOD: A total of 589 CT examinations performed at a single center between March 2010 and August 2017 were retrospectively included. The DL algorithm was designed to segment AVC and to quantify AVC volume, and Agatston score was calculated using attenuation values. Manually measured AVC volume and Agatston score were used as ground truth. To validate AVC segmentation performance, the Dice coefficient was calculated. For observer performance testing, four radiologists determined AVC grade in two reading rounds. The diagnostic performance of DL-measured AVC volume and Agaston score for classifying severe AVC was compared with that of each reader's assessment. RESULTS: After applying the DL algorithm, the Dice coefficient score was 0.807. In patients with AVC, accuracy of DL-measured AVC volume for AVC grading was 97.0 % with area under the curve (AUC) of 0.964 (95 % confidence interval [CI] 0.923-1) in the test set, which was better than the radiologist readers (accuracy 69.7 %-91.9 %, AUC 0.762-0.923) with manually measured AVC volume as ground truth. When manually measured AVC Agatston score was used as ground truth, accuracy of DL-measured AVC Agatston score for AVC grading was 92.9 % with AUC of 0.933 (95 % CI 0.885-0.981) in the test set, which was also better than the radiologist readers (accuracy 77.8-89.9 %, AUC 0.791-0.903). CONCLUSIONS: DL-based automated AVC quantification may be comparable with manual measurements. The diagnostic performance of the DL-measured AVC volume and Agatston score for classification of severe AVC outperforms radiologist readers.

Differentiation of recurrent glioblastoma from radiation necrosis using diffusion radiomics with machine learning model development and external validation.

The purpose of this study was to establish a high-performing radiomics strategy with machine learning from conventional and diffusion MRI to differentiate recurrent glioblastoma (GBM) from radiation necrosis (RN) after concurrent chemoradiotherapy (CCRT) or radiotherapy. Eighty-six patients with GBM were enrolled in the training set after they underwent CCRT or radiotherapy and presented with new or enlarging contrast enhancement within the radiation field on follow-up MRI. A diagnosis was established either pathologically or clinicoradiologically (63 recurrent GBM and 23 RN). Another 41 patients (23 recurrent GBM and 18 RN) from a different institution were enrolled in the test set. Conventional MRI sequences (T2-weighted and postcontrast T1-weighted images) and ADC were analyzed to extract 263 radiomic features. After feature selection, various machine learning models with oversampling methods were trained with combinations of MRI sequences and subsequently validated in the test set. In the independent test set, the model using ADC sequence showed the best diagnostic performance, with an AUC, accuracy, sensitivity, specificity of 0.80, 78%, 66.7%, and 87%, respectively. In conclusion, the radiomics models models using other MRI sequences showed AUCs ranging from 0.65 to 0.66 in the test set. The diffusion radiomics may be helpful in differentiating recurrent GBM from RN..

Diffusion tensor and postcontrast T1-weighted imaging radiomics to differentiate the epidermal growth factor receptor mutation status of brain metastases from non-small cell lung cancer.

PURPOSE: To assess whether the radiomic features of diffusion tensor imaging (DTI) and conventional postcontrast T1-weighted (T1C) images can differentiate the epidermal growth factor receptor (EGFR) mutation status in brain metastases from non-small cell lung cancer (NSCLC). METHODS: A total of 99 brain metastases in 51 patients who underwent surgery or biopsy with underlying NSCLC and known EGFR mutation statuses (57 from EGFR wild type, 42 from EGFR mutant) were allocated to the training (57 lesions in 31 patients) and test (42 lesions in 20 patients) sets. Radiomic features (n = 526) were extracted from preoperative MR images including T1C and DTI. Radiomics classifiers were constructed by combinations of five feature selectors and four machine learning algorithms. The trained classifiers were validated on the test set, and the classifier performance was assessed by determining the area under the curve (AUC). RESULTS: EGFR mutation status showed an overall discordance rate of 12% between the primary tumors and corresponding brain metastases. The best performing classifier was a combination of the tree-based feature selection and linear discriminant algorithm and 5 features were selected (1 from ADC, 2 from fractional anisotropy, and 2 from T1C images), resulting in an AUC, accuracy, sensitivity, and specificity of 0.73, 78.6%, 81.3%, and 76.9% in the test set, respectively. CONCLUSIONS: Radiomics classifiers integrating multiparametric MRI parameters may have potential in differentiating the EGFR mutation status in brain metastases from NSCLC.

Machine Learning Based Radiomic HPV Phenotyping of Oropharyngeal SCC: A Feasibility Study Using MRI.

OBJECTIVES: To investigate whether a radiomic MRI feature-based prediction model can differentiate oropharyngeal squamous cell carcinoma (SCC) according to the human papillomavirus (HPV) status. STUDY DESIGN: Retrospective cohort study. METHODS: Pretreatment MRI data from 62 consecutive patients with oropharyngeal SCC were retrospectively reviewed, and chronologically allocated to training (n = 43) and test sets (n = 19). Enhancing tumors were semi-automatically segmented on each slice of the postcontrast T1WI to span the entire tumor volume, after registration of T2WI to postcontrast T1WI; 170 radiomic features were extracted from the entire tumor volume. Relevant features were selected and radiomics models were trained using least absolute shrinkage and selection operator (LASSO) logistic regression model with 10-fold cross-validation, after subsampling of training sets using synthetic minority over-sampling technique to mitigate data imbalance. The selected features, weighted by their respective coefficients, were combined linearly to yield a radiomics score. The diagnostic performance of the radiomic score was evaluated using the area under the receiver operating characteristic curve (AUC). RESULTS: Six radiomic features, which revealed strong association with HPV status of oropharyngeal SCC, were selected using LASSO. The radiomics model yielded excellent performance on the training set (AUC, 0.982 [95% CI, 0.942-1.000]) and moderate performance on the test set (AUC, 0.744 [95% CI, 0.496-0.991]) for differentiating oropharyngeal SCC according to HPV status. CONCLUSIONS: Radiomics-based MRI phenotyping differentiates oropharyngeal SCC according to HPV status, and thus, is a potential imaging biomarker. LEVEL OF EVIDENCE: 3 Laryngoscope, 131:E851-E856, 2021.

Radiomics risk score may be a potential imaging biomarker for predicting survival in isocitrate dehydrogenase wild-type lower-grade gliomas.

OBJECTIVES: Isocitrate dehydrogenase wild-type (IDHwt) lower-grade gliomas of histologic grades II and III follow heterogeneous clinical outcomes, which necessitates risk stratification. We aimed to evaluate whether radiomics from MRI would allow prediction of overall survival in patients with IDHwt lower-grade gliomas and to investigate the added prognostic value of radiomics over clinical features. METHODS: Preoperative MRIs of 117 patients with IDHwt lower-grade gliomas from January 2007 to February 2018 were retrospectively analyzed. The external validation cohort consisted of 33 patients from The Cancer Genome Atlas. A total of 182 radiomic features were extracted. Radiomics risk scores (RRSs) for overall survival were derived from the least absolute shrinkage and selection operator (LASSO) and elastic net. Multivariable Cox regression analyses, including clinical features and RRSs, were performed. The integrated areas under the receiver operating characteristic curves (iAUCs) from models with and without RRSs were calculated for comparisons. The prognostic value of RRS was assessed in the validation cohort. RESULTS: The RRS derived from LASSO and elastic net independently predicted survival with hazard ratios of 9.479 (95% confidence interval [CI], 3.220-27.847) and 6.148 (95% CI, 3.009-12.563), respectively. Those RRSs enhanced model performance for predicting overall survival (iAUC increased to 0.780-0.797 from 0.726), which was externally validated. The RRSs stratified IDHwt lower-grade gliomas in the validation cohort with significantly different survival. CONCLUSION: Radiomics has the potential for noninvasive risk stratification and can improve prediction of overall survival in patients with IDHwt lower-grade gliomas when integrated with clinical features. KEY POINTS: • Isocitrate dehydrogenase wild-type lower-grade gliomas with histologic grades II and III follow heterogeneous clinical outcomes, which necessitates further risk stratification. • Radiomics risk scores derived from MRI independently predict survival even after incorporating strong clinical prognostic features (hazard ratios 6.148-9.479). • Radiomics risk scores derived from MRI have the potential to improve survival prediction when added to clinical features (integrated areas under the receiver operating characteristic curves increased from 0.726 to 0.780-0.797).

Robust performance of deep learning for distinguishing glioblastoma from single brain metastasis using radiomic features: model development and validation.

We evaluated the diagnostic performance and generalizability of traditional machine learning and deep learning models for distinguishing glioblastoma from single brain metastasis using radiomics. The training and external validation cohorts comprised 166 (109 glioblastomas and 57 metastases) and 82 (50 glioblastomas and 32 metastases) patients, respectively. Two-hundred-and-sixty-five radiomic features were extracted from semiautomatically segmented regions on contrast-enhancing and peritumoral T2 hyperintense masks and used as input data. For each of a deep neural network (DNN) and seven traditional machine learning classifiers combined with one of five feature selection methods, hyperparameters were optimized through tenfold cross-validation in the training cohort. The diagnostic performance of the optimized models and two neuroradiologists was tested in the validation cohort for distinguishing glioblastoma from metastasis. In the external validation, DNN showed the highest diagnostic performance, with an area under receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy of 0.956 (95% confidence interval [CI], 0.918-0.990), 90.6% (95% CI, 80.5-100), 88.0% (95% CI, 79.0-97.0), and 89.0% (95% CI, 82.3-95.8), respectively, compared to the best-performing traditional machine learning model (adaptive boosting combined with tree-based feature selection; AUC, 0.890 (95% CI, 0.823-0.947)) and human readers (AUC, 0.774 [95% CI, 0.685-0.852] and 0.904 [95% CI, 0.852-0.951]). The results demonstrated deep learning using radiomic features can be useful for distinguishing glioblastoma from metastasis with good generalizability.

Deep Learning for Automated Detection of Cyst and Tumors of the Jaw in Panoramic Radiographs.

Patients with odontogenic cysts and tumors may have to undergo serious surgery unless the lesion is properly detected at the early stage. The purpose of this study is to evaluate the diagnostic performance of the real-time object detecting deep convolutional neural network You Only Look Once (YOLO) v2-a deep learning algorithm that can both detect and classify an object at the same time-on panoramic radiographs. In this study, 1602 lesions on panoramic radiographs taken from 2010 to 2019 at Yonsei University Dental Hospital were selected as a database. Images were classified and labeled into four categories: dentigerous cysts, odontogenic keratocyst, ameloblastoma, and no cyst. Comparative analysis among three groups (YOLO, oral and maxillofacial surgeons, and general practitioners) was done in terms of precision, recall, accuracy, and F1 score. While YOLO ranked highest among the three groups (precision = 0.707, recall = 0.680), the performance differences between the machine and clinicians were statistically insignificant. The results of this study indicate the usefulness of auto-detecting convolutional networks in certain pathology detection and thus morbidity prevention in the field of oral and maxillofacial surgery.

Development of real-time motion verification system using in-room optical images for respiratory-gated radiotherapy.

Phase-based respiratory-gated radiotherapy relies on the reproducibility of patient breathing during the treatment. To monitor the positional reproducibility of patient breathing against a 4D CT simulation, we developed a real-time motion verification system (RMVS) using an optical tracking technology. The system in the treatment room was integrated with a real-time position management system. To test the system, an anthropomorphic phantom that was mounted on a motion platform moved on a programmed breathing pattern and then underwent a 4D CT simulation with RPM. The phase-resolved anterior surface lines were extracted from the 4D CT data to constitute 4D reference lines. In the treatment room, three infrared reflective markers were attached on the superior, middle, and inferior parts of the phantom along with the body midline and then RMVS could track those markers using an optical camera system. The real-time phase information extracted from RPM was delivered to RMVS via in-house network software. Thus, the real-time anterior-posterior positions of the markers were simultaneously compared with the 4D reference lines. The technical feasibility of RMVS was evaluated by repeating the above procedure under several scenarios such as ideal case (with identical motion parameters between simulation and treatment), cycle change, baseline shift, displacement change, and breathing type changes (abdominal or chest breathing). The system capability for operating under irregular breathing was also investigated using real patient data. The evaluation results showed that RMVS has a competence to detect phase-matching errors between patient's motion during the treatment and 4D CT simulation. Thus, we concluded that RMVS could be used as an online quality assurance tool for phase-based gating treatments.

Development of an optical-based image guidance system: technique detecting external markers behind a full facemask.

PURPOSE: Optical image-guided systems (e.g., AlignRT, frameless SonArray, ExacTrac) have been used with advantages of avoiding excessive radiation exposure and real-time patient monitoring. Although these systems showed proven accuracy, they need to modify a full facemask for patients with H&N cancer and brain tumor. We developed an optical-based guidance system to manage interfractional and intrafractional setup errors by tracking external markers behind a full facemask. METHODS: Infra-red (IR) reflecting markers were attached on the face of a head phantom and then the phantom was immobilized by a full face thermoplastic mask. A stereo camera system consisting of two CCD cameras was mounted on the inferior wall of treatment room. The stereo camera system was calibrated to reconstruct 3D coordinates of multiple markers with respect to the isocenter using the direct linear transform (DLT) algorithm. The real-time position of the phantom was acquired, through the stereo camera system, by detecting the IR markers behind the full facemask. The detection errors with respect to the reference positions of planning CT images were calculated in six degrees of freedom (6-DOF) by a rigid-body registration technique. RESULTS: The calibration accuracy of the system was in submillimeter (0.33 mm +/- 0.27 mm), which was comparable to others. The mean distance between each of marker positions of optical images and planning CT images was 0.50 mm +/- 0.67 mm. The maximum deviations of 6-DOF registration were less than 1 mm and 1 degrees for the couch translation and rotation, respectively. CONCLUSIONS: The developed system showed the accuracy and consistency comparable to the commercial optical guided systems, while allowing us to simultaneously immobilize patients with a full face thermoplastic mask.

Quasi-breath-hold technique using personalized audio-visual biofeedback for respiratory motion management in radiotherapy.

PURPOSE: To introduce a respiratory motion management technique, so called quasi-breath-hold (QBH) technique and evaluate its feasibility. As a hybrid technique combining free-breathing-based gating (denoted as gating for convenience) and breath-hold (BH), the QBH is designed to overcome typical limitations existing in either one such as phase-shift, residual motion, complexity, and discomfort. METHODS: The QBH is realized using an audio-visual biofeedback system (AVBFS) and a respiratory motion management program (RMMP). The AVBFS, consisting of two infra-red stereo cameras and a head mounted display, monitors respiratory motion and provides dynamic feedback to patients. The RMMP establishes a personalized respiration model based on deep free breathing. The model is further processed to generate a QBH model by inserting a short breath-hold period into the end point of the-end-of-expiration phase. Then the patient is guided to follow the QBH model through the AVBFS. A simulation study with ten volunteers was performed to evaluate the feasibility of the proposed technique. In the simulation, an in-house developed macro program automatically controlled the QBH procedure to virtually deliver an intensity modulated radiation therapy (IMRT) plan. For each volunteer subject, three QBH maneuvers with different breath-hold times of 3, 5, and 7s (denoted as QBH3s, QBH5s, and QBH7s, respectively) and a conventional gating maneuver with 30% duty cycle (for comparison purpose) were applied. External respiration motion signals obtained during the gating window were analyzed to obtain mean absolute error (MAE) between the measured and guiding curve, mean absolute deviation (MAD) of the measured curve, and an inverse uncertainty time histogram (IUTH). RESULTS: Every volunteer successfully performed all of the four maneuvers (1 gating and 3 QBH patterns). The average treatment times were 466.8, 452.3, and 430.8 s for the QBH3s, QBH5s, and QBH7s, respectively, compared to 530.4 s for the gating technique. The mean absolute errors between measured and guiding curve during the gating window were 0.9 +/- 0.7, 0.8 +/- 0.6, 0.7 +/- 0.6, and 0.6 +/- 0.7 mm for the gating, QBH3s, QBH5s, and QBH7s, respectively. The mean absolute deviations of the measured curve during the gating window were 0.7 +/- 0.7, 0.5 +/- 0.5, 0.5 +/- 0.4, and 0.5 +/- 0.6 mm for the gating, QBH3s, QBH5s, and QBH7s, respectively. In the analysis of the IUTH during the gating window, the QBH simulations showed similar (QBH3s) or less (QBH5s and QBH7s) motion uncertainties compared to the gating simulation. CONCLUSIONS: The proposed QBH technique with personalized audio-visual biofeedback was feasible for respiratory motion management. It showed equivalent or less motion uncertainty and shorter treatment time than the conventional free-breathing-based gating technique did. The technique is expected to optimally compromise between patient comfort and treatment efficiency.