Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

Radiology Radiology

Deep learning for predicting epidermal growth factor receptor mutations of non-small cell lung cancer on PET/CT images.

In Quantitative imaging in medicine and surgery

BACKGROUND : Predicting the mutation status of the epidermal growth factor receptor (EGFR) gene based on an integrated positron emission tomography/computed tomography (PET/CT) image of non-small cell lung cancer (NSCLC) is a noninvasive, low-cost method which is valuable for targeted therapy. Although deep learning has been very successful in robotic vision, it is still challenging to predict gene mutations in PET/CT-derived studies because of the small amount of medical data and the different parameters of PET/CT devices.

METHODS : We used the advanced EfficientNet-V2 model to predict the EGFR mutation based on fused PET/CT images. First, we extracted 3-dimensional (3D) pulmonary nodules from PET and CT as regions of interest (ROIs). We then fused each single PET and CT image. The network model was used to predict the mutation status of lung nodules by the new data after fusion, and the model was weighted adaptively. The EfficientNet-V2 model used multiple channels to represent nodules comprehensively.

RESULTS : We trained the EfficientNet-V2 model through our PET/CT fusion algorithm using a dataset of 150 patients. The prediction accuracy of EGFR and non-EGFR mutations was 86.25% in the training dataset, and the accuracy rate was 81.92% in the validation set.

CONCLUSIONS : Combined with experiments, the demonstrated PET/CT fusion algorithm outperformed radiomics methods in predicting EGFR and non-EGFR mutations in NSCLC.

Xiao Zhenghui, Cai Haihua, Wang Yue, Cui Ruixue, Huo Li, Lee Elaine Yuen-Phin, Liang Ying, Li Xiaomeng, Hu Zhanli, Chen Long, Zhang Na

2023-Mar-01

EfficientNet-V2 Model, epidermal growth factor receptor (EGFR), positron emission tomography/computed tomography (PET/CT)

General General

Detecting obstructive coronary artery disease with machine learning: rest-only gated single photon emission computed tomography myocardial perfusion imaging combined with coronary artery calcium score and cardiovascular risk factors.

In Quantitative imaging in medicine and surgery

BACKGROUND : The rest-only single photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) has low diagnostic performance for obstructive coronary artery disease (CAD). Coronary artery calcium score (CACS) is strongly associated with obstructive CAD. The aim of this study was to investigate the performance of rest-only gated SPECT MPI combined with CACS and cardiovascular risk factors in diagnosing obstructive CAD through machine learning (ML).

METHODS : We enrolled 253 suspected CAD patients who underwent the 1-stop rest-only SPECT MPI and computed tomography (CT) scan due to stress test-related contraindications. Myocardial perfusion and wall motion were assessed using quantitative perfusion SPECT + quantitative gated SPECT (QPS + QGS) automated quantification software. The Agatston algorithm was used to calculate CACS. The clinical data of patients, including cardiovascular risk factors, were collected. Based on feature selection and clinical experience, 8 factors were identified as modeling variables. Subsequently, patients were divided randomly into 2 groups: the training (70%) and test (30%) groups. The performance of 8 supervised ML algorithms was evaluated in the training and test groups.

RESULTS : Obstructive CAD was diagnosed by coronary angiography in 94 (37.2%, 94/253) patients. In the training group, the area under the receiver operator characteristic (ROC) curve (AUC) of the random forest was the highest, and the AUCs of Logistic, extreme gradient boosting (XGBoost), support vector machine (SVM), and adaptive boosting (AdaBoost) were all above 0.9. In the test group, the AUC of recursive partitioning and regression trees (Rpart) was the highest (0.911). Rpart and Naïve Bayes had the highest accuracy (0.840). Rpart had a sensitivity and specificity of 0.851 and 0.821, respectively; Naïve Bayes had a sensitivity and specificity of 0.809 and 0.893, respectively. Next was Logistic, with an accuracy of 0.827, a sensitivity of 0.872, and a specificity of 0.750. The random forest and XGBoost algorithms also had high accuracy, which was 0.813 for each algorithm.

CONCLUSIONS : Rest-only SPECT MPI combined with CACS and cardiovascular risk factors using an ML algorithm to detect obstructive CAD is feasible. Among the algorithms validated in the test group, Rpart, Naïve Bayes, XGBoost, Logistic, and random forest are all highly accurate for diagnosing obstructive CAD. The application of ML in resting MPI and CACS may be used for screening obstructive CAD.

Liu Bao, Yu Wenji, Zhang Feifei, Shi Yunmei, Yang Le, Jiang Qi, Wang Yufeng, Wang Yuetao

2023-Mar-01

Machine learning (ML), coronary artery calcium score (CACS), coronary artery disease (CAD), myocardial perfusion imaging (MPI), single photon emission computed tomography (SPECT)

oncology Oncology

Machine learning for predicting accuracy of lung and liver tumor motion tracking using radiomic features.

In Quantitative imaging in medicine and surgery

BACKGROUND : Internal tumor motion is commonly predicted using external respiratory signals. However, the internal/external correlation is complex and patient-specific. The purpose of this study was to develop various models based on the radiomic features of computed tomography (CT) images to predict the accuracy of tumor motion tracking using external surrogates and to find accurate and reliable tracking algorithms.

METHODS : Images obtained from a total of 108 and 71 patients pathologically diagnosed with lung and liver cancers, respectively, were examined. Real-time position monitoring motion was fitted to tumor motion, and samples with fitting errors greater than 2 mm were considered positive. Radiomic features were extracted from internal target volumes of average intensity projections, and cross-validation least absolute shrinkage and selection operator (LassoCV) was used to conduct feature selection. Based on the radiomic features, a total of 26 separate models (13 for the lung and 13 for the liver) were trained and tested. Area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were used to assess performance. Relative standard deviation was used to assess stability.

RESULTS : Thirty-three and 22 radiomic features were selected for the lung and liver, respectively. For the lung, the AUC varied from 0.848 (decision tree) to 0.941 [support vector classifier (SVC), logistic regression]; sensitivity varied from 0.723 (extreme gradient boosting) to 0.848 [linear support vector classifier (linearSVC)]; specificity varied from 0.834 (gaussian naive bayes) to 0.936 [multilayer perceptron (MLP), wide and deep (W&D)]; and MLP and W&D had better performance and stability than the median. For the liver, the AUC varied from 0.677 [light gradient boosting machine (Light)] to 0.892 (logistic regression); sensitivity varied from 0.717 (W&D) to 0.862 (MLP); specificity varied from 0.566 (Light) to 0.829 (linearSVC); and logistic regression, MLP, and SVC had better performance and stability than the median.

CONCLUSIONS : Respiratory-sensitive radiomic features extracted from CT images of lung and liver tumors were proved to contain sufficient information to establish an external/internal motion relationship. We developed a rapid and accurate method based on radiomics to classify the accuracy of monitoring a patient's external surface for lung and liver tumor tracking. Several machine learning algorithms-in particular, MLP-demonstrated excellent classification performance and stability.

Li Guangjun, Zhang Xiangyu, Song Xinyu, Duan Lian, Wang Guangyu, Xiao Qing, Li Jing, Liang Lan, Bai Long, Bai Sen

2023-Mar-01

Tumor motion, machine learning, radiomics, respiratory motion, tumor tracking

Radiology Radiology

Dark blood T2-weighted imaging of the human heart with AI-assisted compressed sensing: a patient cohort study.

In Quantitative imaging in medicine and surgery

BACKGROUND : Dark blood T2-weighted (DB-T2W) imaging is widely used to evaluate myocardial edema in myocarditis and inflammatory cardiomyopathy. However, this technique is sensitive to arrhythmia, tachycardia, and cardiac and respiratory motion due to the long scan time with multiple breath-holds. The application of artificial intelligence (AI)-assisted compressed sensing (ACS) has facilitated significant progress in accelerating medical imaging. However, the effect of DB-T2W imaging on ACS has not been elucidated. This study aimed to examine the effects of ACS on the image quality of single-shot and multi-shot DB-T2W imaging of edema.

METHODS : Thirty-three patients were included in this study and received DB-T2W imaging with ACS, including single-shot acquisition (SS-ACS) and multi-shot acquisition (MS-ACS). The resulting images were compared with those of the conventional multi-shot DB-T2W imaging with parallel imaging (MS-PI). Quantitative assessments of the signal-to-noise ratio (SNR), tissue contrast ratio (CR), and contrast-to-noise ratio (CNR) were performed. Three radiologists independently evaluated the overall image quality, blood nulling, free wall of the left ventricle, free wall of the right ventricle, and interventricular septum using a 5-point Likert scale.

RESULTS : The total scan time of the DB-T2W imaging with ACS was significantly reduced compared to the conventional parallel imaging [number of heartbeats (SS-ACS:MS-ACS:MS-PI) =19:63:99; P<0.001]. The SNRmyocardium and CNRblood-myocardium of MS-ACS and SS-ACS were higher than those of MS-PI (all P values <0.01). Furthermore, the CRblood-myocardium of SS-ACS was also higher than that of MS-PI (P<0.01). There were significant differences in overall image quality, blood nulling, left ventricle free wall visibility, and septum visibility between the MS-PI, MS-ACS, and SS-ACS protocols (all P values <0.05). Moreover, blood in the heart was better nulled using SS-ACS (P<0.01).

CONCLUSIONS : The ACS method shortens the scan time of DB-T2W imaging and achieves comparable or even better image quality compared to the PI method. Moreover, DB-T2W imaging using the ACS method can reduce the number of breath-holds to 1 with single-shot acquisition.

Yan Xianghu, Ran Lingping, Zou Lixian, Luo Yi, Yang Zhaoxia, Zhang Shiyu, Zhang Shuheng, Xu Jian, Huang Lu, Xia Liming

2023-Mar-01

Artificial intelligence (AI), MRI, T2-weighted imaging, compressed sensing, dark blood

General General

A systematic review of the modelling of patient arrivals in emergency departments.

In Quantitative imaging in medicine and surgery

BACKGROUND : Accident and Emergency Department (AED) is the frontline of providing emergency care in a hospital and research focusing on improving decision-makings and service level around AED has been driving a rising number of attentions in recent years. A retrospective review among the published papers shows that related research can be classified according to six planning modules: demand forecasting, days-off scheduling, shift scheduling, line-of-work construction, task assignment and staff assignment. As patient arrivals demand forecasts enable smooth AED operational planning and help decision-making, this article conducted a systematic review on the statistical modelling approaches aimed at predicting the volume of AED patients' arrival.

METHODS : We carried out a systematic review of AED patient arrivals prediction studies from 2004 to 2021. The Medline, ScienceDirect, and Scopus databases were searched. A two-step screening process was carried out based on the title and abstract or full text, and 35 of 1,677 articles were selected. Our methods and results follow the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines. We categorise AED methods for modelling patient arrivals into four main classes: regression, time series, artificial intelligence and time series regression. Choice of prediction model, selection of factors and model performance are compared. Finally, we discuss the advantages and limitations of the models and suggest future research directions.

RESULTS : A total of 1,677 papers that fulfilled the initial searching criteria was obtained from the three databases. Based on the first exclusion criteria, 1,603 articles were eliminated. The remaining 74 full text articles were evaluated based on the second exclusion criteria. Finally, 35 articles were selected for full review. We find that the use of artificial intelligence-based model has risen in recent years, from the view of predictive model selection. The calendar-based factors are most commonly used compared with other types of dependent variables, from the view of dependent variable selection.

CONCLUSIONS : All AEDs are inherently different and different covariables may have different effects on patient arrivals. Certain factors may play a key role in one AED but not others. Based on results of meta-analysis, when modelling patient arrivals, it is essential to understand the actual AED situation and carefully select relevant dominating factors and the most suitable modelling method. Local calibration is also important to ensure good estimates.

Jiang Shancheng, Liu Qize, Ding Beichen

2023-Mar-01

AED patient arrivals, PRISMA, healthcare data analytics, predicting models

Ophthalmology Ophthalmology

Deep learning-based image analysis of eyelid morphology in thyroid-associated ophthalmopathy.

In Quantitative imaging in medicine and surgery

BACKGROUND : We aimed to propose a deep learning-based approach to automatically measure eyelid morphology in patients with thyroid-associated ophthalmopathy (TAO).

METHODS : This prospective study consecutively included 74 eyes of patients with TAO and 74 eyes of healthy volunteers visiting the ophthalmology department in a tertiary hospital. Patients diagnosed as TAO and healthy volunteers who were age- and gender-matched met the eligibility criteria for recruitment. Facial images were taken under the same light conditions. Comprehensive eyelid morphological parameters, such as palpebral fissure (PF) length, margin reflex distance (MRD), eyelid retraction distance, eyelid length, scleral area, and mid-pupil lid distance (MPLD), were automatically calculated using our deep learning-based analysis system. MRD1 and 2 were manually measured. Bland-Altman plots and intraclass correlation coefficients (ICCs) were performed to assess the agreement between automatic and manual measurements of MRDs. The asymmetry of the eyelid contour was analyzed using the temporal: nasal ratio of the MPLD. All eyelid features were compared between TAO eyes and control eyes using the independent samples t-test.

RESULTS : A strong agreement between automatic and manual measurement was indicated. Biases of MRDs in TAO eyes and control eyes ranged from -0.01 mm [95% limits of agreement (LoA): -0.64 to 0.63 mm] to 0.09 mm (LoA: -0.46 to 0.63 mm). ICCs ranged from 0.932 to 0.980 (P<0.001). Eyelid features were significantly different in TAO eyes and control eyes, including MRD1 (4.82±1.59 vs. 2.99±0.81 mm; P<0.001), MRD2 (5.89±1.16 vs. 5.47±0.73 mm; P=0.009), upper eyelid length (UEL) (27.73±4.49 vs. 25.42±4.35 mm; P=0.002), lower eyelid length (LEL) (31.51±4.59 vs. 26.34±4.72 mm; P<0.001), and total scleral area (SATOTAL) (96.14±34.38 vs. 56.91±14.97 mm2; P<0.001). The MPLDs at all angles showed significant differences in the 2 groups of eyes (P=0.008 at temporal 180°; P<0.001 at other angles). The greatest temporal-nasal asymmetry appeared at 75° apart from the midline in TAO eyes.

CONCLUSIONS : Our proposed system allowed automatic, comprehensive, and objective measurement of eyelid morphology by only using facial images, which has potential application prospects in TAO. Future work with a large sample of patients that contains different TAO subsets is warranted.

Shao Ji, Huang Xingru, Gao Tao, Cao Jing, Wang Yaqi, Zhang Qianni, Lou Lixia, Ye Juan

2023-Mar-01

Thyroid-associated ophthalmopathy (TAO), automatic measurement, deep learning, eyelid morphology, facial images