Search

IntroductionPreoperative evaluation of the mitotic index (MI) of gastrointestinal stromal tumors (GISTs) represents the basis of individualized treatment of patients. However, the accuracy of conventional preoperative imaging methods is limited. The aim of this study was to develop a predictive model based on multiparametric MRI for preoperative MI prediction.MethodsA total of 112 patients who were pathologically diagnosed with GIST were enrolled in this study. The dataset was subdivided into the development (n = 81) and test (n = 31) sets based on the time of diagnosis. With the use of T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) map, a convolutional neural network (CNN)-based classifier was developed for MI prediction, which used a hybrid approach based on 2D tumor images and radiomics features from 3D tumor shape. The trained model was tested on an internal test set. Then, the hybrid model was comprehensively tested and compared with the conventional ResNet, shape radiomics classifier, and age plus diameter classifier.ResultsThe hybrid model showed good MI prediction ability at the image level; the area under the receiver operating characteristic curve (AUROC), area under the precision–recall curve (AUPRC), and accuracy in the test set were 0.947 (95% confidence interval [CI]: 0.927–0.968), 0.964 (95% CI: 0.930–0.978), and 90.8 (95% CI: 88.0–93.0), respectively. With the average probabilities from multiple samples per patient, good performance was also achieved at the patient level, with AUROC, AUPRC, and accuracy of 0.930 (95% CI: 0.828–1.000), 0.941 (95% CI: 0.792–1.000), and 93.6% (95% CI: 79.3–98.2) in the test set, respectively.DiscussionThe deep learning-based hybrid model demonstrated the potential to be a good tool for the operative and non-invasive prediction of MI in GIST patients.

BackgroundRapid serial visual presentation (RSVP) has become a popular target detection method by decoding electroencephalography (EEG) signals, owing to its sensitivity and effectiveness. Most current research on EEG-based RSVP tasks focused on feature extraction algorithms developed to deal with the non-stationarity and low signal-to-noise ratio (SNR) of EEG signals. However, these algorithms cannot handle the problem of no event-related potentials (ERP) component or miniature ERP components caused by the attention lapses of human vision in abnormal conditions. The fusion of human-computer vision can obtain complementary information, making it a promising way to become an efficient and general way to detect objects, especially in attention lapses.MethodsDynamic probability integration (DPI) was proposed in this study to fuse human vision and computer vision. A novel basic probability assignment (BPA) method was included, which can fully consider the classification capabilities of different heterogeneous information sources for targets and non-targets and constructs the detection performance model for the weight generation based on classification capabilities. Furthermore, a spatial-temporal hybrid common spatial pattern-principal component analysis (STHCP) algorithm was designed to decode EEG signals in the RSVP task. It is a simple and effective method of distinguishing target and non-target using spatial-temporal features.ResultsA nighttime vehicle detection based on the RSVP task was performed to evaluate the performance of DPI and STHCP, which is one of the conditions of attention lapses because of its decrease in visual information. The average AUC of DPI was 0.912 ± 0.041 and increased by 11.5, 5.2, 3.4, and 1.7% compared with human vision, computer vision, naive Bayesian fusion, and dynamic belief fusion (DBF), respectively. A higher average balanced accuracy of 0.845 ± 0.052 was also achieved using DPI, representing that DPI has the balanced detection capacity of target and non-target. Moreover, STHCP obtained the highest AUC of 0.818 ± 0.06 compared with the other two baseline methods and increased by 15.4 and 23.4%.ConclusionExperimental results indicated that the average AUC and balanced accuracy of the proposed fusion method were higher than individual detection methods used for fusion, as well as two excellent fusion methods. It is a promising way to improve detection performance in RSVP tasks, even in abnormal conditions.

BackgroundRelapse is the major cause of mortality in patients with resected endometrial cancer (EC). There is an urgent need for a feasible method to identify patients with high risk of relapse.PurposeTo develop a multi-parameter magnetic resonance imaging (MRI) radiomics-based nomogram model to predict 5-year progression-free survival (PFS) in EC.MethodsFor this retrospective study, 202 patients with EC followed up for at least 5 years after hysterectomy. A radiomics signature was extracted from T2-weighted imaging (T2WI), apparent diffusion coefficient (ADC) and a dynamic contrast-enhanced three-dimensional volumetric interpolated breath-hold examination (3D-VIBE). The radiomics score (RS) was calculated based on the least absolute shrinkage and selection operator (LASSO) regression. We have developed a radiomics based nomogram model (ModelN) incorporating the RS and clinical and conventional MR (cMR) risk factors. The performance was evaluated by the receiver operating characteristic curve (ROC), calibration curve and decision curve analysis (DCA).ResultsThe ModelN demonstrated a good calibration and satisfactory discrimination, with a mean area under the curve (AUC) of 0.840 and 0.958 in the training and test cohorts, respectively. In comparison with clinical prediction model (ModelC), the discrimination ability of ModelN showed an improvement with P < 0.001 for the training cohort and P=0.032 for the test cohort. Compared to the radiomics prediction model (ModelR), ModelN discrimination ability showed an improvement for the training cohort with P = 0.021, with no statistically significant difference in the test cohort (P = 0.106). Calibration curves suggested a good fit for probability (Hosmer–Lemeshow test, P = 0.610 and P = 0.956 for the training and test cohorts, respectively).ConclusionThis multi-parameter nomogram model incorporating clinical and cMR findings is a valid method to predict 5-year PFS in patients with EC.

Brain–computer interface (BCI) has developed rapidly over the past two decades, mainly due to advancements in machine learning. Subjects must learn to modulate their brain activities to ensure a successful BCI. Feedback training is a practical approach to this learning process; however, the commonly used classifier-dependent approaches have inherent limitations such as the need for calibration and a lack of continuous feedback over long periods of time. This paper proposes an online data visualization feedback protocol that intuitively reflects the EEG distribution in Riemannian geometry in real time. Rather than learning a hyperplane, the Riemannian geometry formulation allows iterative learning of prototypical covariance matrices that are translated into visualized feedback through diffusion map process. Ten subjects were recruited for MI-BCI (motor imagery-BCI) training experiments. The subjects learned to modulate their sensorimotor rhythm to centralize the points within one category and to separate points belonging to different categories. The results show favorable overall training effects in terms of the class distinctiveness and EEG feature discriminancy over a 3-day training with 30% learners. A steadily increased class distinctiveness in the last three sessions suggests that the advanced training protocol is effective. The optimal frequency band was consistent during the 3-day training, and the difference between subjects with good or low MI-BCI performance could be clearly observed. We believe that the proposed feedback protocol has promising application prospect.