Mean Radial Error Research Articles

Does the FARNet neural network algorithm accurately identify Posteroanterior cephalometric landmarks?

BackgroundWe explored whether the feature aggregation and refinement network (FARNet) algorithm accurately identified posteroanterior (PA) cephalometric landmarks.MethodsWe identified 47 landmarks on 1,431 PA cephalograms of which 1,177 were used for training, 117 for validation, and 137 for testing. A FARNet-based artificial intelligence (AI) algorithm automatically detected the landmarks. Model effectiveness was calculated by deriving the mean radial error (MRE) and the successful detection rates (SDRs) within 2, 2.5, 3, and 4 mm. The Mann-Whitney U test was performed on the Euclidean differences between repeated manual identifications and AI trials. The direction in differences was analyzed, and whether differences moved in the same or opposite directions relative to ground truth on both the x and y-axis.ResultsThe AI system (web-based CranioCatch annotation software (Eskişehir, Turkey)) identified 47 anatomical landmarks in PA cephalograms. The right gonion SDRs were the highest, thus 96.4, 97.8, 100, and 100% within 2, 2.5, 3, and 4 mm, respectively. The right gonion MRE was 0.94 ± 0.53 mm. The right condylon SDRs were the lowest, thus 32.8, 45.3, 54.0, and 67.9% within the same thresholds. The right condylon MRE was 3.31 ± 2.25 mm. The AI model’s reliability and accuracy were similar to a human expert’s. AI was better at four skeleton points than the expert, whereas the expert was better at one skeletal and seven dental points (P < 0.05). Most of the points exhibited significant deviations along the y-axis. Compared to ground truth, most of the points in AI and the second trial showed opposite movement on the x-axis and the same on the y-axis.ConclusionsThe FARNet algorithm streamlined orthodontic diagnosis.

Cone-beam CT landmark detection for measuring basal bone width: a retrospective validation study

BackgroundAccurate assessment of basal bone width is essential for distinguishing individuals with normal occlusion from patients with maxillary transverse deficiency who may require maxillary expansion. Herein, we evaluated the effectiveness of a deep learning (DL) model in measuring landmarks of basal bone width and assessed the consistency of automated measurements compared to manual measurements.MethodsBased on the U-Net algorithm, a coarse-to-fine DL model was developed and trained using 80 cone-beam computed tomography (CBCT) images. The model’s prediction capabilities were validated on 10 CBCT scans and tested on an additional 34. To evaluate the performance of the DL model, its measurements were compared with those taken manually by one junior orthodontist using the concordance correlation coefficient (CCC).ResultsIt took approximately 1.5 s for the DL model to perform the measurement task in only CBCT images. This framework showed a mean radial error of 1.22 ± 1.93 mm and achieved successful detection rates of 71.34%, 81.37%, 86.77%, and 91.18% in the 2.0-, 2.5-, 3.0-, and 4.0-mm ranges, respectively. The CCCs (95% confidence interval) of the maxillary basal bone width and mandibular basal bone width distance between the DL model and manual measurement for the 34 cases were 0.96 (0.94–0.97) and 0.98 (0.97–0.99), respectively.ConclusionThe novel DL framework developed in this study improved the diagnostic accuracy of the individual assessment of maxillary width. These results emphasize the potential applicability of this framework as a computer-aided diagnostic tool in orthodontic practice.

Automatic localization of anatomical landmarks in head cine fluoroscopy images via deep learning.

Fluoroscopy guided interventions (FGIs) pose a risk of prolonged radiation exposure; personalized patient dosimetry is necessary to improve patient safety during these procedures. However, current FGIs systems do not capture the precise exposure regions of the patient, making it challenging to perform patient-procedure-specific dosimetry. Thus, there is a pressing need to develop approaches to extract and use this information to enable personalized radiation dosimetry for interventional procedures. To propose a deep learning (DL) approach for the automatic localization of 3D anatomical landmarks on randomly collimated and magnified 2D head fluoroscopy images. The model was developed with datasets comprising 800000 pseudo 2D synthetic images (mixture of vessel-enhanced and non-enhancement), each with 55 annotated anatomical landmarks (two are landmarks for eye lenses), generated from 135 retrospectively collected head computed tomography (CT) volumetric data. Before training, dynamic random cropping was performed to mimic the varied field-size collimation in FGI procedures. Gaussian-distributed additive noise was applied to each individual image to enhance the robustness of the DL model in handling image degradation that may occur during clinical image acquisition in a clinical environment. The model was trained with 629370 synthetic images for approximately 275000 iterations and evaluated against a synthetic image test set and a clinical fluoroscopy test set. The model shows good performance in estimating in- and out-of-image landmark positions and shows feasibility to instantiate the skull shape. The model successfully detected 96.4% and 92.5% 2D and 3D landmarks, respectively, within a 10mm error on synthetic test images. It demonstrated an average of 3.6±2.3mm mean radial error and successfully detected 96.8% 2D landmarks within 10mm error on clinical fluoroscopy images. Our deep-learning model successfully localizes anatomical landmarks and estimates the gross shape of skull structures from collimated 2D projection views. This method may help identify the exposure region required for patient-specific organ dosimetry in FGIs procedures.

Automatic soft-tissue analysis on orthodontic frontal and lateral facial photographs based on deep learning.

To establish the automatic soft-tissue analysis model based on deep learning that performs landmark detection and measurement calculations on orthodontic facial photographs to achieve a more comprehensive quantitative evaluation of soft tissues. A total of 578 frontal photographs and 450 lateral photographs of orthodontic patients were collected to construct datasets. All images were manually annotated by two orthodontists with 43 frontal-image landmarks and 17 lateral-image landmarks. Automatic landmark detection models were established, which consisted of a high-resolution network, a feature fusion module based on depthwise separable convolution, and a prediction model based on pixel shuffle. Ten measurements for frontal images and eight measurements for lateral images were defined. Test sets were used to evaluate the model performance, respectively. The mean radial error of landmarks and measurement error were calculated and statistically analysed to evaluate their reliability. The mean radial error was 14.44 ± 17.20 pixels for the landmarks in the frontal images and 13.48 ± 17.12 pixels for the landmarks in the lateral images. There was no statistically significant difference between the model prediction and manual annotation measurements except for the mid facial-lower facial height index. A total of 14 measurements had a high consistency. Based on deep learning, we established automatic soft-tissue analysis models for orthodontic facial photographs that can automatically detect 43 frontal-image landmarks and 17 lateral-image landmarks while performing comprehensive soft-tissue measurements. The models can assist orthodontists in efficient and accurate quantitative soft-tissue evaluation for clinical application.

Can artificial intelligence-driven cephalometric analysis replace manual tracing? A systematic review and meta-analysis.

This systematic review and meta-analysis aimed to investigate the accuracy and efficiency of artificial intelligence (AI)-driven automated landmark detection for cephalometric analysis on two-dimensional (2D) lateral cephalograms and three-dimensional (3D) cone-beam computed tomographic (CBCT) images. An electronic search was conducted in the following databases: PubMed, Web of Science, Embase, and grey literature with search timeline extending up to January 2024. Studies that employed AI for 2D or 3D cephalometric landmark detection were included. The selection of studies, data extraction, and quality assessment of the included studies were performed independently by two reviewers. The risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. A meta-analysis was conducted to evaluate the accuracy of the 2D landmarks identification based on both mean radial error and standard error. Following the removal of duplicates, title and abstract screening, and full-text reading, 34 publications were selected. Amongst these, 27 studies evaluated the accuracy of AI-driven automated landmarking on 2D lateral cephalograms, while 7 studies involved 3D-CBCT images. A meta-analysis, based on the success detection rate of landmark placement on 2D images, revealed that the error was below the clinically acceptable threshold of 2mm (1.39mm; 95% confidence interval: 0.85-1.92mm). For 3D images, meta-analysis could not be conducted due to significant heterogeneity amongst the study designs. However, qualitative synthesis indicated that the mean error of landmark detection on 3D images ranged from 1.0 to 5.8mm. Both automated 2D and 3D landmarking proved to be time-efficient, taking less than 1min. Most studies exhibited a high risk of bias in data selection (n = 27) and reference standard (n = 29). The performance of AI-driven cephalometric landmark detection on both 2D cephalograms and 3D-CBCT images showed potential in terms of accuracy and time efficiency. However, the generalizability and robustness of these AI systems could benefit from further improvement. PROSPERO: CRD42022328800.

Microelectrode Recording During Deep Brain Stimulation Does Not Consistently Represent Lead Trajectory.

Long-term outcomes in deep brain stimulation (DBS) depend on accuracy of lead placement. Microelectrode recording (MER) is a long-used adjunct to leverage neurophysiological information to confirm satisfactory trajectory of implanted electrodes. The goal of this study was to evaluate the consistency in which electrodes are placed in sampled microelectrode trajectories. This is a retrospective study using intraoperative computed tomography to measure final electrode deviation from MER probe placement during the DBS insertion targeting subthalamic nucleus. Fifteen patients had 29 DBS leads placed using MER assistance. Radial distance between the probe and the lead were measured for each patient using intraoperative imaging. In addition, the preoperative target to final lead error was measured in 14 patients undergoing subthalamic nucleus implants without the use of MER and compared with the 15 patients in which MER was used as an adjunct. There was no significant difference in the mean radial target error (1.2 vs 1.0 mm, P = .156) when comparing the leads placed with or without MER assistance, respectively. The mean difference in final position of microelectrode compared with DBS lead was 0.9 ± 0.1 (range 0.4-2.0 mm). Of all MER-assisted electrodes placed, 44.8% (13) of electrode final positions radially deviated 1.0 mm or more from the MER probe. Electrode placement may deviate significantly from MER trajectories. Given the concern that physiological data may not be representative of the final electrode trajectory, surgeons should consider using intraoperative imaging or other adjunctive techniques during DBS to confirm accuracy and satisfactory trajectory of DBS leads.

Accuracy, Repeatability, and Reproducibility of a Hand-Held Structured-Light 3D Scanner across Multi-Site Settings in Lower Limb Prosthetics.

The aim of this work was to assess the accuracy, repeatability, and reproducibility of a hand-held, structured-light 3D scanner (EINScan Pro 2X Plus with High Definition Prime Pack, SHINING 3D Tech. Co., Ltd., Hangzhou, China), to support its potential use in multi-site settings on lower limb prosthetics. Four limb models with different shapes were fabricated and scanned with a metrological 3D scanner (EINScan Laser FreeScan 5X, SHINING 3D Tech. Co., Ltd., Hangzhou, China) by a professional operator (OP0). Limb models were then mailed to three sites where two operators (OP1, OP2) scanned them using their own structured-light 3D scanner (same model). OP1 scanned limb models twice (OP1-A, OP1-B). OP0, OP1-A, and OP2 scans were compared for accuracy, OP1-A and OP1-B for repeatability, and OP1-A and OP2 for reproducibility. Among all comparisons, the mean radial error was <0.25 mm, mean angular error was <4°, and root mean square error of the radial distance was <1 mm. Moreover, limits of agreement were <3.5% for perimeters and volumes. By comparing these results with respect to clinically-relevant thresholds and to the literature available on other 3D scanners, we conclude that the EINScan Pro 2X Plus 3D Scanner with High Definition Prime Pack has good accuracy, repeatability, and reproducibility, supporting its use in multi-site settings.

Three-Dimensional Structure Light Robot-Assisted Frameless Stereotactic Brain Biopsy.

To assess the feasibility, accuracy, and safety of 3-dimensional (3D) structure light robot-assisted frameless stereotactic brain biopsy. Five consecutive patients (3 males, 2 females) were included in this study. The patients' clinical, imaging, and histological data were analyzed, and all patients received a 3D structure light robot-assisted frameless stereotactic brain biopsy. The raw and/or analyzed data of the study are available from the corresponding author. The statistical results showed a mean age of 59.6 years (range 40-70 years), a mean target depth of 60.9 mm (range 53.5-65.8 mm), a mean radial error of 1.2 ± 0.7 mm (mean ± SD), a mean depth error of 0.7 ± 0.3 mm, and a mean absolute tip error of 1.5 ± 0.6 mm. The calculated Pearson product-moment correlation coefficient ( r = 0.23) revealed no correlation between target depth and absolute tip error. All biopsy needles were placed in line with the planned trajectory successfully, and diagnostic specimens were harvested in all cases. Histopathological analysis revealed lymphoma (2 cases), lung adenocarcinoma (1 case), glioblastoma multiforme (1 case), and oligodendroglioma (1 case). Surface registration using the 3D structure light technique is fast and precise because of the achievable million-scale point cloud data of the head and face. 3D structure light robot-assisted frameless stereotactic brain biopsy is feasible, accurate, and safe.

A new EEG neurofeedback training approach in sports: the effects function-specific instruction of Mu rhythm and visuomotor skill performance.

Achieving optimal visuomotor performance in precision sports relies on maintaining an optimal psychological state during motor preparation. To uncover the optimal psychological state, extensive EEG studies have established a link between the Mu rhythm (8-13 Hz at Cz) and cognitive resource allocation during visuomotor tasks (i.e., golf or shooting). In addition, the new approach in EEG neurofeedback training (NFT), called the function-specific instruction (FSI) approach, for sports involves providing function-directed verbal instructions to assist individuals to control specific EEG parameters and align them with targeted brain activity features. While this approach was initially hypothesized to aid individuals in attaining a particular mental state during NFT, the impact of EEG-NFT involving Mu rhythm on visuomotor performance, especially when contrasting the traditional instruction (TI) approach with the FSI approach, underscores the necessity for additional exploration. Hence, the objective of this study is to investigate the impact of the FSI approach on modulating Mu rhythm through EEG-NFT in the context of visuomotor performance. Thirty novice participants were recruited and divided into three groups: function-specific instruction (FSI, four females, six males; mean age = 27.00 ± 7.13), traditional instruction (TI, five females, five males; mean age = 27.00 ± 3.88), and sham control (SC, five females, five males; mean age = 27.80 ± 5.34). These groups engaged in a single-session EEG-NFT and performed golf putting tasks both before and after the EEG-NFT. The results showed that within the FSI group, single-session NFT with augmented Mu power led to a significant decrease in putting performance (p = 0.013). Furthermore, we noted a marginal significance indicating a slight increase in Mu power and a reduction in the subjective sensation of action control following EEG-NFT (p = 0.119). While there was a positive correlation between Mu power and mean radial error in golf putting performance (p = 0.043), it is important to interpret this relationship cautiously in the context of reduced accuracy in golf putting. The findings emphasize the necessity for extended investigation to attain a more profound comprehension of the nuanced significance of Mu power in visuomotor performance. The study highlights the potential effectiveness of the FSI approach in EEG-NFT and in enhancing visuomotor performance, but it also emphasizes the potential impact of skill level and attentional control, particularly in complex visuomotor tasks.

Landmark annotation through feature combinations: a comparative study on cephalometric images with in-depth analysis of model's explainability.

The objectives of this study are to explore and evaluate the automation of anatomical landmark localization in cephalometric images using machine learning techniques, with a focus on feature extraction and combinations, contextual analysis, and model interpretability through Shapley Additive exPlanations (SHAP) values. We conducted extensive experimentation on a private dataset of 300 lateral cephalograms to thoroughly study the annotation results obtained using pixel feature descriptors including raw pixel, gradient magnitude, gradient direction, and histogram-oriented gradient (HOG) values. The study includes evaluation and comparison of these feature descriptions calculated at different contexts namely local, pyramid, and global. The feature descriptor obtained using individual combinations is used to discern between landmark and nonlandmark pixels using classification method. Additionally, this study addresses the opacity of LGBM ensemble tree models across landmarks, introducing SHAP values to enhance interpretability. The performance of feature combinations was assessed using metrics like mean radial error, standard deviation, success detection rate (SDR) (2 mm), and test time. Remarkably, among all the combinations explored, both the HOG and gradient direction operations demonstrated significant performance across all context combinations. At the contextual level, the global texture outperformed the others, although it came with the trade-off of increased test time. The HOG in the local context emerged as the top performer with an SDR of 75.84% compared to others. The presented analysis enhances the understanding of the significance of different features and their combinations in the realm of landmark annotation but also paves the way for further exploration of landmark-specific feature combination methods, facilitated by explainability.

Automatic detection of posterior superior alveolar artery in dental cone-beam CT images using a deeply supervised multi-scale 3D network.

This study aimed to develop a robust and accurate deep learning network for detecting the posterior superior alveolar artery (PSAA) in dental cone-beam CT (CBCT) images, focusing on the precise localization of the centre pixel as a critical centreline pixel. PSAA locations were manually labelled on dental CBCT data from 150 subjects. The left maxillary sinus images were horizontally flipped. In total, 300 datasets were created. Six different deep learning networks were trained, including 3D U-Net, deeply supervised 3D U-Net (3D U-Net DS), multi-scale deeply supervised 3D U-Net (3D U-Net MSDS), 3D Attention U-Net, 3D V-Net, and 3D Dense U-Net. The performance evaluation involved predicting the centre pixel of the PSAA. This was assessed using mean absolute error (MAE), mean radial error (MRE), and successful detection rate (SDR). The 3D U-Net MSDS achieved the best prediction performance among the tested networks, with an MAE measurement of 0.696 ± 1.552 mm and MRE of 1.101 ± 2.270 mm. In comparison, the 3D U-Net showed the lowest performance. The 3D U-Net MSDS demonstrated a SDR of 95% within a 2 mm MAE. This was a significantly higher result than other networks that achieved a detection rate of over 80%. This study presents a robust deep learning network for accurate PSAA detection in dental CBCT images, emphasizing precise centre pixel localization. The method achieves high accuracy in locating small vessels, such as the PSAA, and has the potential to enhance detection accuracy and efficiency, thus impacting oral and maxillofacial surgery planning and decision-making.

Visualization modality for augmented reality guidance of in-depth tumour enucleation procedures.

Recent research studies reported that the employment of wearable augmented reality (AR) systems such as head-mounted displays for the in situ visualisation of ultrasound (US) images can improve the outcomes of US-guided biopsies through reduced procedure completion times and improved accuracy. Here, the authors continue in the direction of recent developments and present the first AR system for guiding an in-depth tumour enucleation procedure under US guidance. The system features an innovative visualisation modality with cutting trajectories that 'sink' into the tissue according to the depth reached by the electric scalpel, tracked in real-time, and a virtual-to-virtual alignment between the scalpel's tip and the trajectory. The system has high accuracy in estimating the scalpel's tip position (mean depth error of 0.4 mm and mean radial error of 1.34 mm). Furthermore, we demonstrated with a preliminary user study that our system allowed us to successfully guide an in-depth tumour enucleation procedure (i.e. preserving the safety margin around the lesion).

Comparison of 2 Robotic Systems for Pediatric Stereoelectroencephalography Implantation

Stereoelectroencephalography (SEEG) remains critical in guiding epilepsy surgery. Robot-assisted techniques have shown promise in improving SEEG implantation outcomes but have not been directly compared. In this single-institution series, we compared ROSA and Stealth AutoGuide robots in pediatric SEEG implantation. We retrospectively reviewed 21 sequential pediatric SEEG implantations consisting of 6 ROSA and 15 AutoGuide procedures. We determined mean operative time, time per electrode, root mean square (RMS) registration error, and surgical complications. Three-dimensional radial distances were calculated between each electrode's measured entry and target points with respective errors from the planned trajectory line. Mean overall/per electrode operating time was 73.5/7.5 minutes for ROSA and 126.1/10.9 minutes for AutoGuide (P= 0.030 overall, P= 0.082 per electrode). Mean RMS registration error was 0.77 mm (0.55-0.93 mm) for ROSA and 0.6 mm (0.2-1.0 mm) for AutoGuide (P= 0.26). No procedures experienced complications. The mean radial (entry point error was 1.23 ± 0.11 mm for ROSA and 2.65 ± 0.12 mm for AutoGuide (P < 0.001), while the mean radial target point error was 1.86 ± 0.15 mm for ROSA and 3.25 ± 0.16 mm for AutoGuide (P < 0.001). Overall operative time was greater for AutoGuide procedures, although there was no statistically significant difference in time per electrode. Both systems are highly accurate with no significant RMS error difference. While the ROSA robot yielded significantly lower entry and target point errors, both robots are safe and reliable for deep electrode insertion in pediatric epilepsy.

Automatic axial vertebral rotation estimation on radiographs for adolescent idiopathic scoliosis by deep learning

PurposeAdolescent idiopathic scoliosis (AIS) is a three-dimensional spine deformity governed by lateral curvature and axial vertebral rotation (AVR). Estimating AVR is important for treatment decisions and the prediction of AIS progression. However, manual AVR measurements have intra-observer and inter-observer errors, and therefore this paper proposes an automatic AVR measurement method to address the low precision issue of manual measurement. MethodWe develop an improved feature extraction module for vertebral landmark detection and pedicle segmentation. The improved coordinate convolution layer combined Polarized Self-Attention mechanism is applied in the feature extraction module to help the High-resolution Network to extract coordinate information. Based on vertebral landmark detection and pedicle segmentation, we propose an automatic AVR measurement algorithm to estimate the AVR. ResultsThe mean radial error (MRE) of vertebral landmark detection is 2.70 mm pixels, and the dice coefficient of pedicle segmentation is 72.45%. Compared with the original model, the MRE decreases by 2.13 mm, and the dice coefficient improves by 4.61%. For the AVR measurement results, the testing set contains 37 spine radiographs, including 481 vertebrae and 962 AVR measurements (481 vertebrae by two observers). The average measurement error is lower than 5°, which is within the standard of clinical measurement error. ConclusionThe results demonstrate that the proposed network performs well in vertebral landmark detection and pedicle segmentation, and the proposed AVR measurement is adopted for clinic diagnosis of AIS. SignificanceOur method achieves automatic AVR measurement, reducing the error introduced by manual measurement and improving the efficiency of orthopedists.

Stereo-EEG-guided network modulation for psychiatric disorders: Surgical considerations

BackgroundDeep brain stimulation (DBS) and other neuromodulatory techniques are being increasingly utilized to treat refractory neurologic and psychiatric disorders. Objective/Hypothesis: To better understand the circuit-level pathophysiology of treatment-resistant depression (TRD) and treat the network-level dysfunction inherent to this challenging disorder, we adopted an approach of inpatient intracranial monitoring borrowed from the epilepsy surgery field. MethodsWe implanted 3 patients with 4 DBS leads (bilateral pair in both the ventral capsule/ventral striatum and subcallosal cingulate) and 10 stereo-electroencephalography (sEEG) electrodes targeting depression-relevant network regions. For surgical planning, we used an interactive, holographic visualization platform to appreciate the 3D anatomy and connectivity. In the initial surgery, we placed the DBS leads and sEEG electrodes using robotic stereotaxy. Subjects were then admitted to an inpatient monitoring unit for depression-specific neurophysiological assessments. Following these investigations, subjects returned to the OR to remove the sEEG electrodes and internalize the DBS leads to implanted pulse generators. ResultsIntraoperative testing revealed positive valence responses in all 3 subjects that helped verify targeting. Given the importance of the network-based hypotheses we were testing, we required accurate adherence to the surgical plan (to engage DBS and sEEG targets) and stability of DBS lead rotational position (to ensure that stimulation field estimates of the directional leads used during inpatient monitoring were relevant chronically), both of which we confirmed (mean radial error 1.2±0.9 mm; mean rotation 3.6±2.6°). ConclusionThis novel hybrid sEEG-DBS approach allows detailed study of the neurophysiological substrates of complex neuropsychiatric disorders.

Ceph-Net: automatic detection of cephalometric landmarks on scanned lateral cephalograms from children and adolescents using an attention-based stacked regression network

BackgroundThe success of cephalometric analysis depends on the accurate detection of cephalometric landmarks on scanned lateral cephalograms. However, manual cephalometric analysis is time-consuming and can cause inter- and intra-observer variability. The purpose of this study was to automatically detect cephalometric landmarks on scanned lateral cephalograms with low contrast and resolution using an attention-based stacked regression network (Ceph-Net).MethodsThe main body of Ceph-Net compromised stacked fully convolutional networks (FCN) which progressively refined the detection of cephalometric landmarks on each FCN. By embedding dual attention and multi-path convolution modules in Ceph-Net, the network learned local and global context and semantic relationships between cephalometric landmarks. Additionally, the intermediate deep supervision in each FCN further boosted the training stability and the detection performance of cephalometric landmarks.ResultsCeph-Net showed a superior detection performance in mean radial error and successful detection rate, including accuracy improvements in cephalometric landmark detection located in low-contrast soft tissues compared with other detection networks. Moreover, Ceph-Net presented superior detection performance on the test dataset split by age from 8 to 16 years old.ConclusionsCeph-Net demonstrated an automatic and superior detection of cephalometric landmarks by successfully learning local and global context and semantic relationships between cephalometric landmarks in scanned lateral cephalograms with low contrast and resolutions.

FDGR-Net: Feature Decouple and Gated Recalibration Network for medical image landmark detection

Medical image landmark detection can automatically determine the pre-set position coordinates in medical images, so as to further assist disease diagnosis and treatment planning. Using Convolutional Neural Network (CNN) to estimate the heatmap of the landmark has become the mainstream method. However, some landmarks are not obvious and there are local similarities between different landmarks in the medical images. Existing methods usually directly map the image features to all the landmarks, without considering these two challenges. In this paper, we propose a Feature Decouple and Gated Recalibration Network (FDGR-Net) for the medical image landmark detection. The FDGR-Net first establishes independent pathways to decouple all landmark features into single landmark features. In this process, a dynamic multi-kernel convolution is designed as the decouple basic component to improve the feature robustness of the not obvious landmarks. Then, a novel Gated Feature Recalibration Module (GFRM) is proposed, which includes a retention gate and an elimination gate. The retention gate can further strengthen the information of the target landmark and the elimination gate can eliminate the interference of non-target landmarks. The proposed method is validated on one public dataset and two in-house datasets, and its performance is superior to the existing advanced method. The Mean Radial Error (MRE) of the proposed method can reach 3.04, 5.51 and 1.45 in the Knee dataset, Hand dataset and Head dataset, respectively.

Automatic identification of posteroanterior cephalometric landmarks using a novel deep learning algorithm: a comparative study with human experts

This study aimed to propose a fully automatic posteroanterior (PA) cephalometric landmark identification model using deep learning algorithms and compare its accuracy and reliability with those of expert human examiners. In total, 1032 PA cephalometric images were used for model training and validation. Two human expert examiners independently and manually identified 19 landmarks on 82 test set images. Similarly, the constructed artificial intelligence (AI) algorithm automatically identified the landmarks on the images. The mean radial error (MRE) and successful detection rate (SDR) were calculated to evaluate the performance of the model. The performance of the model was comparable with that of the examiners. The MRE of the model was 1.87 ± 1.53 mm, and the SDR was 34.7%, 67.5%, and 91.5% within error ranges of < 1.0, < 2.0, and < 4.0 mm, respectively. The sphenoid points and mastoid processes had the lowest MRE and highest SDR in auto-identification; the condyle points had the highest MRE and lowest SDR. Comparable with human examiners, the fully automatic PA cephalometric landmark identification model showed promising accuracy and reliability and can help clinicians perform cephalometric analysis more efficiently while saving time and effort. Future advancements in AI could further improve the model accuracy and efficiency.

An automatic cephalometric landmark detection method based on heatmap regression and Monte Carlo dropout.

Cephalometric analysis plays an important role in orthodontic diagnosis and treatment planning. It depends on the detection of multiple landmarks, while the process is time-consuming and tedious. Although some deep learning-based automatic landmark detection algorithms have achieved excellent performance, most of them adopt multi-stage models increasing the complexity and detection time. Meanwhile, few studies focused on the uncertainty of detection results, thereby ignoring its significant clinical value. In this paper, we propose a novel approach based on heatmap regression for landmark detection, which can achieve competitive accuracy and good robustness with only one step. Furthermore, by applying Monte Carlo dropout to a U-shaped convolutional neural network, we can obtain not only the coordinate of each landmark but also the corresponding simple uncertainty, so that doctors can pay more attention to those landmarks with higher uncertainty. The evaluation results showed the mean radial error is 1.39±1.06mm and the average successful detection rate is 79.65%, 97.22% within 2mm, 4mm for the IEEE ISBI2015 Test Dataset 1, the indicators for the IEEE ISBI2015 Test Dataset 2 are 1.33±0.93mm, 80.05% and 97.53%, respectively. Our method has the potential to become an assistant tool in clinical practice. Automatic and accurate detection with uncertainty analysis is expected to help guide the doctor's judgment.

Estimating marksmanship performance during walking while maintaining weapon aim

Marksmanship performance while moving is a critical skill among tactical athletes due to the high demands of their occupational duties. Qualifications for dynamic marksmanship performance are not standardized across tactical athlete groups, which may limit comprehensive assessment of tactical athlete performance for situational awareness and adaptability to an unpredictable environment. Although static marksmanship performance provides foundational information on skills and level of ability, research is lacking on factors that influence dynamic marksmanship performance to best prepare tactical athletes for duties. The purpose of this study was to identify whether static marksmanship performance, speed of movement, load carriage, and biomechanical factors while ‘shooting on the move’ influenced dynamic marksmanship performance. Twenty-four male tactical athletes (22 active-duty Army Soldiers, two civilian SWAT operators; age: 23.83 ± 5.47 years; height: 1.80 ± 0.08 m; weight: 81.04 ± 7.87 kg) participated; final analyses did not include data from the two civilian operators to maintain sample homogeneity. Tactical athletes completed static and dynamic (‘shoot on the move’) marksmanship tasks under three load conditions: (1) no load (NL), (2) half kit (HK) of 11.34 kg, and (3) full kit (KIT) of 22.68 kg. Dynamic marksmanship was completed under three speed conditions: (1) self-selected slow speed, (2) standard speed, and (3) self-selected fast speed. Hip, knee, and ankle kinematics were collected via wireless inertial measurement units. Spatiotemporal parameters were collected via optical detection system. Marksmanship performance (accuracy) was collected via open-air acoustic target scoring and mean radial error (MRE) was calculated for both static and dynamic marksmanship tasks. Linear mixed-effects models were fit with dynamic MRE as the outcome variable with fixed effects of static MRE, load condition, speed condition, kinematics, and spatiotemporal parameters, adjusting for body mass. Alpha level was set a priori at p ≤ 0.10. The final statistical model included fixed effects of static MRE, load condition, speed condition, and time spent in double limb support. Static MRE (p < 0.01) and time spent in double limb support (p = 0.01) were significant factors. For each 1 cm increase in static MRE there was a 0.66 cm increase in dynamic MRE. For every 1% increase in time spent in double limb support while ‘shooting on the move’ there was a 0.13 cm increase in dynamic MRE. Findings from this study highlight that tactical athletes who have larger static stance MRE and spend a longer time in double limb support during a gait cycle exhibit an increase in MRE during ‘shoot on the move’ trials. Overall, dynamic shooting accuracy is not affected by lower extremity joint angles, load carriage, or speed of movement. Although strong relationships are known between gait speed, load, and lower extremity kinematics, the differences in tactical gait compared to normal gait and multi-task paradigm that likely favors marksmanship accuracy seem to present novel movement characteristics unique to occupational gait. Further investigation is warranted to identify other potential factors that may improve or worsen dynamic marksmanship performance.