BACKGROUND:An accurate assessment of metastatic lymph nodes (MLNs) in head and neck magnetic resonance imaging (MRIs) is crucial in nasopharyngeal carcinoma (NPC) staging and treatment planning. As a fundamental step in computer-aided diagnosis systems, automated MLNs segmentation remains challenging due to three key issues: the small size, blurred margins and large localization variations among patients, which often lead to over- and under-segmentation. PURPOSE:To address these challenges, the Small- and Large-scale Information Collaborative Unet (SLSIC-Unet) has been proposed. This network is specifically designed to aggregate small-scale information, modality-specific features (MSFs) and the spatial relationships across MRI slices. METHOD:The proposed SLSIC-Unet incorporates three dedicated strategies corresponding to the aforementioned challenges. First, to prevent small MLNs from being obscured by background information, a small- and large-scale information collaborative framework (SLSICF) is proposed. This framework utilizes dual interactive pathways: magnified image patches are processed to aggregate small-scale information, preventing the loss of small-area MLN representations, while raw images are simultaneously analyzed to extract large-scale contextual information, ensuring performance for general MLNs. Second, to delineate the fuzzy margins of MLNs, the SLSIC-Unet leverages complementary information across modalities by extracting both modality-specific features (MSFs) and multi-modality-fusion features. Finally, the spatial feature context module (SFCM) is designed to aggregate the spatial relationships of MRI slices, utilizing the spatial distribution tendency of MLNs which mostly exhibit proximity to blood vessels. RESULTS:Evaluated on a cohort of 663 NPC cases, the proposed model achieved valuable segmentation performance, with mean values of 0.801 ± 0.098 for normalized surface distance (NSD), 0.823 ± 0.082 for dice similarity coefficient (DSC), and 0.706 ± 0.107 for intersection over union (IoU) on the test set. Comparative analysis revealed that the SLSIC-Unet significantly outperformed (p < 0.05) existing state-of-the-art methods in MLNs segmentation. CONCLUSION:The SLSIC-Unet is capable of accurately segmenting MLNs, and can provide technical support for assisting in Nstaging of nasopharyngeal carcinoma.

Optimal treatment decision-making is critical to the long-term survival rate of patients in clinical management of cancer. Traditional treatment decision-making models typically reach treatment decisions indirectly by estimating the various risks to the beneficial outcome of patients, which limits the interpretability of treatment effects and the performance of the models. We propose SurvS(survival supervision), a novel interpretable survival benefit analytics framework based on the real-coded genetic algorithm, to integrate individual treatment effects that directly impact long-term survival benefits. SurvS can not only construct binary (beneficial and nonbeneficial), but also construct ternary (beneficial, insensitive, and detrimental) decision-making models, allowing the identification of all three possible outcomes in a single model. The framework generates personalized model scores by integrating weighted clinical features and real-valued cutoff thresholds, optimized to reflect survival outcome differences. The robust performances of SurvS were demonstrated in decision-making tasks of two clinical cancer treatment scenarios: induction chemotherapy for nasopharyngeal carcinoma and adjuvant chemotherapy for rectal cancer. SurvS outperformed traditional methods and maintained robust performance in the independent validation cohort. By enabling survival benefit-supervised optimization and interpretable model construction, reported SurvS offers a powerful tool for personalized cancer treatment planning with the potential to improve treatment efficacy and reduce overtreatment. The code is open source and available at: https://github.com/odindis/SurvS .

Automatic segmentation of liver metastases with fuzzy boundary from 3D images is a challenging task, and the issue of segmenting fuzzy boundary remains unsolved in existing deep learning-based methods. In this work, we propose the multi-dimensional attention (MDA) module incorporated into U-shaped network to enable precise and automatic segmentation of liver metastases with fuzzy boundary. MDA focuses on the depth, height, width, and channel dimensions of medical images, which can better enhance the model's perception of primary, intermediate, and advanced features. A total of 278 cases of liver metastases data were used, with 228 cases for 5fold cross-validation, and 50 cases for testing. The comparative results showed that our method achieved the best overall performance, with DSC of 0.780 +/- 0.215, HD95 of 29.155 +/- 43.873 (pixels), and NSD (tau = 2 mm) of 0.836 +/- 0.226. The experimental results show that MDA enhances the capability of U-shaped model to segment the fuzzy boundary of liver metastases by focusing on channel features and capturing extensive contextual information across multiple projection space dimensions. The delineation performance analysis also shows that 46.25 % of the segmentation results were considered satisfactory and required no revisions. This study can provide a new technical support for the segmentation of 3D liver metastases.

Background: Tumor tissues exhibit significant spatial heterogeneity, which affects proliferation rate, invasiveness, and sensitivity to drugs and radiotherapy. Traditional radiomics typically quantifies intratumoral heterogeneity using global whole-tumor features, but does not explicitly model spatial subregions or localized heterogeneity patterns. Therefore, this study aimed to develop and validate a T2-weighted magnetic resonance imaging (MRI) habitat-based multi-regional spatial interaction (MSI) model to non-invasively characterize intratumoral spatial heterogeneity and improve mortality prediction in nasopharyngeal carcinoma (NPC). Methods: We retrospectively enrolled 1,297 NPC patients for death risk prediction. Using the K-Means clustering algorithm, tumor regions were decomposed into four biologically distinct habitat subregions. A MSI matrix was constructed to systematically extract 45 MSI features characterizing tumor heterogeneity. Feature analysis revealed significant differences between high-and low-risk groups in Boundary Volume Class 3-4 (t-test, P=0.024) and Volume Class 4 (t-test, P=0.011). Due to severe imbalance in death samples, a stratified sampling strategy was employed to create 10 balanced training subsets and one test set. After feature selection, a decision tree model was trained on each balanced training subset, and the final prediction for the test set was obtained through a weighted ensemble of the ten models based on their respective training areas under the curve (AUCs). In addition, a conventional radiomics model was constructed in parallel as a comparative baseline. Results: The experimental results demonstrated that the decision tree model constructed using the MSI features achieved the best predictive performance on the test set [AUC: 0.79 (95% confidence interval: 0.69-0.88), accuracy: 0.75, recall: 0.80], outperforming other machine learning models such as random forest and logistic regression. Further evaluation using calibration curves and decision curve analysis demonstrated that the model achieves well-calibrated probability estimation and robust discrimination, providing reliable predictions with a higher net clinical benefit across different risk thresholds, and exhibiting superior prognostic performance compared with conventional radiomics models. Conclusions: Therefore, the MSI features derived from MRI-based habitat analysis enable quantitative assessment of the intratumoral spatial heterogeneity of NPC. As an interpretable biomarker framework, this approach overcomes the limitations of conventional radiomics in which heterogeneity is summarized in a global, non-spatial manner. The resulting model provides an innovative imaging-based solution for prognostic evaluation in NPC and holds potential clinical value for risk stratification and the development of individualized treatment strategies.

BACKGROUND:Accurate classification of high-risk and low-risk thymomas is critical for guiding treatment strategies and assessing prognosis. Thymoma is the most common primary tumor of the anterior mediastinum. However, previous studies have limitations in comprehensively utilizing imaging data, particularly in combining radiomics and deep learning (DL) features for preoperative classification. PURPOSE:This study aimed to develop and validate a comprehensive model based on computed tomography (CT) imaging data (CSRT, Clinical Semantic, Radiomics, and Vision Transformer) to enhance the accuracy of preoperative high-risk and low-risk classification of thymomas and evaluate its application in non-invasive diagnosis. METHODS:This retrospective study included 360 patients with pathologically confirmed thymomas from three centers, with 274 cases (Centers A and B) used for model training and 86 cases (Center C) serving as an external validation set. CT images, including non-contrast enhanced CT (NECT) and contrast-enhanced CT (CECT), were used to extract radiomics features and ViT-based DL features, along with calculated Delta features (NECT minus CECT). Clinical semantic features were integrated, and key features were selected using t-tests and least absolute shrinkage and selection operator (LASSO) regression to construct the fusion model. RESULTS:The CSRT model demonstrated excellent performance in the independent validation cohort, achieving an area under the receiver operating characteristic curve (AUC) of 0.835, an accuracy of 77.9%, a sensitivity of 78.4%, and a specificity of 77.1%. Calibration curves indicated high consistency between predictions and actual classifications. Through decision curve analysis, the model exhibited a high net benefit when the threshold probability exceeded 30%, confirming its clinical utility. CONCLUSIONS:The CSRT model effectively differentiates between high-risk and low-risk thymomas preoperatively using CT imaging data. This non-invasive diagnostic tool supports individualized treatment strategies and enhances clinical decision-making, offering significant value for thymoma management by providing reliable classification above clinically relevant risk thresholds.

Esophageal cancer is a highly aggressive gastrointestinal malignancy with a poor prognosis, making accurate prognostic assessment essential for patient care. The performance of the esophageal cancer prognosis model based on conventional radiomics is limited, as it mainly characterizes the spatial features such as texture of the tumor area, and cannot fully describe the complexity of esophageal cancer tumors. Therefore, we incorporate the frequency domain features into radiomics to improve the prognostic ability of esophageal cancer. Three hundred fifteen esophageal cancer patients participated in the death risk prediction experiment, with 80% being the training set and 20% being the testing set. We use fivefold cross validation for training, and fuse the 5 trained models through voting to obtain the final prognostic model for testing. The CatBoost achieved the best performance compared to machine learning methods such as random forests and decision tree. The experimental results showed that the combination of frequency domain and radiomics features achieved the highest performance in death predicting esophageal cancer (accuracy: 0.7423, precision: 0.7470, recall: 0.7375, specification: 0.8030, AUC: 0.8487), which was significantly better than the performance of frequency domain or radiomics features alone. The results of Kaplan-Meier survival analysis validated the performance of our method in death predicting esophageal cancer. The proposed method provides technical support for accurate prognosis of esophageal cancer.

Traditional decision‐making models often focus on risk prediction rather than treatment effects, potentially leading to suboptimal outcomes. This study developed an individual treatment effect (ITE) model to predict the survival benefit of induction chemotherapy (IC) in locoregionally advanced nasopharyngeal carcinoma (LANPC). This study evaluated a bi‐center cohort of 1213 patients with LANPC and devoloped a genetic algorithm‐based ITE model, classifying patients into IC‐beneficial, IC‐ambiguous, and IC‐detrimental groups. Traditional risk‐stratified models based on the least absolute shrinkage and selection operator and AutoML were established for comparison. Overall survival was the primary endpoint. Models’ efficacy was assessed using Kaplan‒Meier survival analysis. Further validation included correlation analysis, restricted cubic spline curves, and distribution pattern evaluation. In the IC‐beneficial group, IC reduced the mortality risk by 68% (adjusted p = .002) and 48% (adjusted p = .029) in the training and validation sets, respectively. Conversely, IC increased mortality risk in the IC‐detrimental group, with adjusted hazard ratios of 2.66 ( p = .031) and 2.11 ( p = .023). No significant survival difference was observed in the IC‐ambiguous group ( p = .285 and .602). For traditional model, while it stratified patients by risk of dead, it performed poorly in guiding IC decisions in the validation cohort ( p > .05 in high‐risk group). Additionally, the ITE score correlated with short‐term treatment efficacy, and exhibited a stronger association with relative hazard change. The ITE model provides an accurate tool for optimizing IC decisions in LANPC, improving survival, short‐term efficacy, and facilitating personalized treatment strategies.

The blurriness of boundaries in medical image target regions hinders further improvement in automatic segmentation accuracy and is a challenging problem. To address this issue, we propose a model called long-distance perceptual UNet (LD-UNet), which has a powerful long-‍distance perception ability and can effectively perceive the semantic context of an entire image. Specifically, LD-UNet utilizes global and local long-distance induction modules, which endow the model with contextual semantic induction capabilities for long-distance feature dependencies. The modules perform long-distance semantic perception at the high and low stages of LD-UNet, respectively, effectively improving the accuracy of local blurred information assessment. We also propose a top-down deep supervision method to enhance the ability of the model to fit data. Then, extensive experiments on four types of tumor data with blurred boundaries are conducted. The dataset includes nasopharyngeal carcinoma, esophageal carcinoma, pancreatic carcinoma, and colorectal carcinoma. The dice similarity coefficient scores obtained by LD-UNet on the four datasets are 73.35%, 85.93%, 70.04%, and 82.71%. Experimental results demonstrate that LD-UNet is more effective in improving the segmentation accuracy of blurred boundary regions than other methods with long-distance perception, such as transformers. Among all models, LD-UNet achieves the best performance. By visualizing the feature dependency field of the models, we further explore the advantages of LD-UNet in segmenting blurred boundaries.

This study utilizes radiomics to explore imaging biomarkers for predicting the recurrence of chronic subdural hematoma (CSDH), aiming to improve the prediction of CSDH recurrence risk. Analyzing CT scans from 64 patients with CSDH, we extracted 107 radiomic features and employed recursive feature elimination (RFE) and the XGBoost algorithm for feature selection and model construction. The feature selection process identified six key imaging biomarkers closely associated with CSDH recurrence: flatness, surface area to volume ratio, energy, run entropy, small area emphasis, and maximum axial diameter. The selection of these imaging biomarkers was based on their significance in predicting CSDH recurrence, revealing deep connections between postoperative variables and recurrence. After feature selection, there was a significant improvement in model performance. The XGBoost model demonstrated the best classification performance, with the average accuracy improving from 46.82

Automatically delineating colorectal cancers with fuzzy boundaries from 3D images is a challenging task, but the problem of fuzzy boundary delineation in existing deep learning-based methods have not been investigated in depth. Here, an encoder-decoder-based U-shaped network (U-Net) based on top-down deep supervision (TdDS) was designed to accurately and automatically delineate the fuzzy boundaries of colorectal cancer. TdDS refines the semantic targets of the upper and lower stages by mapping ground truths that are more consistent with the stage properties than upsampling deep supervision. This stage-specific approach can guide the model to learn a coarse-to-fine delineation process and improve the delineation accuracy of fuzzy boundaries by gradually shrinking the boundaries. Experimental results showed that TdDS is more customizable and plays a role similar to the attentional mechanism, and it can further improve the capability of the model to delineate colorectal cancer contours. A total of 103, 12, and 29 3D pelvic magnetic resonance imaging volumes were used for training, validation, and testing, respectively. The comparative results indicate that the proposed method exhibits the best comprehensive performance, with a dice similarity coefficient (DSC) of 0.805 ± 0.053 and a hausdorff distance (HD) of 9.28 ± 5.14 voxels. In the delineation performance analysis section also showed that 44.49% of the delineation results are satisfactory and do not require revisions. This study can provide new technical support for the delineation of 3D colorectal cancer. Our method is open source, and the code is available athttps://github.com/odindis/TdDS/tree/main.

BACKGROUND:Pancreatic cancer fine delineation in medical images by physicians is a major challenge due to the vast volume of medical images and the variability of patients.PURPOSE:A semi-automatic fine delineation scheme was designed to assist doctors in accurately and quickly delineating the cancer target region to improve the delineation accuracy of pancreatic cancer in computed tomography (CT) images and effectively reduce the workload of doctors.METHODS:A target delineation scheme in image blocks was also designed to provide more information for the deep learning delineation model. The start and end slices of the image block were manually delineated by physicians, and the cancer in the middle slices were accurately segmented using a three-dimensional Res U-Net model. Specifically, the input of the network is the CT image of the image block and the delineation of the cancer in the start and end slices, while the output of the network is the cancer area in the middle slices of the image block. Meanwhile, the model performance of pancreatic cancer delineation and the workload of doctors in different image block sizes were studied.RESULTS:We used 37 3D CT volumes for training, 11 volumes for validating and 11 volumes for testing. The influence of different image block sizes on doctors' workload was compared quantitatively. Experimental results showed that the physician's workload was minimal when the image block size was 5, and all cancer could be accurately delineated. The Dice similarity coefficient was 0.894 ± 0.029, the 95% Hausdorff distance was 3.465 ± 0.710 mm, the normalized surface Dice was 0.969 ± 0.019. By completing the accurate delineation of all the CT images, the speed of the new method is 2.16 times faster than that of manual sketching.CONCLUSION:Our proposed 3D semi-automatic delineative method based on the idea of block prediction could accurately delineate CT images of pancreatic cancer and effectively deal with the challenges of class imbalance, background distractions, and non-rigid geometrical features. This study had a significant advantage in reducing doctors' workload, and was expected to help doctors improve their work efficiency in clinical application.

BackgroundImage registration technology has become an important medical image preprocessing step with the wide application of computer-aided diagnosis technology in various medical image analysis tasks. PurposeWe propose a multiscale feature fusion registration based on deep learning to achieve the accurate registration and fusion of head magnetic resonance imaging (MRI) and solve the problem that general registration methods cannot handle the complex spatial information and position information of head MRI. MethodsOur proposed multiscale feature fusion registration network consists of three sequentially trained modules. The first is an affine registration module that implements affine transformation; the second is to realize non-rigid transformation, a deformable registration module composed of top-down and bottom-up feature fusion subnetworks in parallel; and the third is a deformable registration module that also realizes non-rigid transformation and is composed of two feature fusion subnetworks in series. The network decomposes the deformation field of large displacement into multiple deformation fields of small displacement by multiscale registration and registration, which reduces the difficulty of registration. Moreover, multiscale information in head MRI is learned in a targeted manner, which improves the registration accuracy, by connecting the two feature fusion subnetworks. ResultsWe used 29 3D head MRIs for training and seven volumes for testing and calculated the values of the registration evaluation metrics for the new algorithm to register anterior and posterior lateral pterygoid muscles. The Dice similarity coefficient was 0.745 +/- 0.021, the Hausdorff distance was 3.441 +/- 0.935 mm, the Average surface distance was 0.738 +/- 0.098 mm, and the Standard deviation of the Jacobian matrix was 0.425 +/- 0.043. Our new algorithm achieved a higher registration accuracy compared with state-of-the-art registration methods. ConclusionsOur proposed multiscale feature fusion registration network can realize end-to-end deformable registration of 3D head MRI, which can effectively cope with the characteristics of large deformation displacement and the rich details of head images and provide reliable technical support for the diagnosis and analysis of head diseases.

Background: With the continuous development of machine vision and imaging technology and its application in computer-aided diagnosis, it is clinically important to use computer technology to assist physicians in accurate cataract surgery. The capsulorhexis directly affects the outcome of cataract surgery, therefore, we design a method to automatically determine the virtual boundary of capsulorhexis for cataract surgery planning and tracking in-vivo to help surgeons achieve a more ideal capsulotomy geometry. Methods: In this study, an effective method was proposed to detect and display the location of capsulorhexis in cataract videos in-vivo. The initial step was locating the entire eye area by analyzing the connected components of the mirror reflective points in the image in the cataract surgery video. Then, an operator was designed for ridge edge variation and used to extract pupil edge features. Lastly, circular Hough transform was used to detect the pupillary margin and calculate the boundary between the scleral limbus and the virtual capsulorhexis border in accordance with the pupillary margin and finally displayed it in-vivo during cataract surgery. Results: The method was tested on eight videos of cataract surgery and the results showed that 98.52% accuracy was achieved in the localization of the specular reflection point. We compared the proposed operator with the Sobel, Scharr, Laplace and Canny operators and the results showed that our operator achieved the smallest mean square error with the greatest structural similarity. Conclusions: The analysis demonstrated that the proposed operator outperformed other operators in detection and achieved satisfactory results in the videos of actual cataract surgeries.

Introduction Automatically and accurately delineating the primary nasopharyngeal carcinoma (NPC) tumors in head magnetic resonance imaging (MRI) images is crucial for patient staging and radiotherapy. Inspired by the bilateral symmetry of head and complementary information of different modalities, a multi-modal neural network named BSMM-Net is proposed for NPC segmentation. Methods First, a bilaterally symmetrical patch block (BSP) is used to crop the image and the bilaterally flipped image into patches. BSP can improve the precision of locating NPC lesions and is a simulation of radiologist locating the tumors with the bilateral difference of head in clinical practice. Second, modality-specific and multi-modal fusion features (MSMFFs) are extracted by the proposed MSMFF encoder to fully utilize the complementary information of T1- and T2-weighted MRI. The MSMFFs are then fed into the base decoder to aggregate representative features and precisely delineate the NPC. MSMFF is the output of MSMFF encoder blocks, which consist of six modality-specific networks and one multi-modal fusion network. Except T1 and T2, the other four modalities are generated from T1 and T2 by the BSP and DT modal generate block. Third, the MSMFF decoder with similar structure to the MSMFF encoder is deployed to supervise the encoder during training and assure the validity of the MSMFF from the encoder. Finally, experiments are conducted on the dataset of 7633 samples collected from 745 patients. Results and discussion The global DICE, precision, recall and IoU of the testing set are 0.82, 0.82, 0.86, and 0.72, respectively. The results show that the proposed model is better than the other state-of-the-art methods for NPC segmentation. In clinical diagnosis, the BSMM-Net can give precise delineation of NPC, which can be used to schedule the radiotherapy.

Accurate classification of leukocytes is crucial for the diagnosis of hematologic malignancies, particularly leukemia. However, traditional leukocyte classification methods are time-consuming and subject to subjective interpretation by examiners. To address this issue, we aimed to develop a leukocyte classification system capable of accurately classifying 11 leukocyte classes, which would aid radiologists in diagnosing leukemia. Our proposed two-stage classification scheme involved a multi-model fusion based on ResNet for rough leukocyte classification, which focused on shape features, followed by fine-grained leukocyte classification using support vector machine for lymphocytes based on texture features. Our dataset consisted of 11,102 microscopic leukocyte images of 11 classes. Our proposed method achieved accurate leukocyte subtype classification with high levels of accuracy, sensitivity, specificity, and precision of 97.03 ± 0.05, 96.76 ± 0.05, 99.65 ± 0.05, and 96.54 ± 0.05, respectively, in the test set. The experimental results demonstrate that the leukocyte classification model based on multi-model fusion can effectively classify 11 leukocyte classes, providing valuable technical support for enhancing the performance of hematology analyzers.

The camera function of a smartphone can be used to quantitatively detect urine parameters anytime, anywhere. However, the color captured by different cameras in different environments is different. A method for color correction is proposed for a urine test strip image collected using a smartphone. In this method, the color correction model is based on the color information of the urine test strip, as well as the ambient light and camera parameters. Conv-TabNet, which can focus on each feature parameter, was designed to correct the color of the color blocks of the urine test strip. The color correction experiment was carried out in eight light sources on four mobile phones. The experimental results show that the mean absolute error of the new method is as low as 2.8±1.8, and the CIEDE2000 color difference is 1.5±1.5. The corrected color is almost consistent with the standard color by visual evaluation. This method can provide a technology for the quantitative detection of urine test strips anytime and anywhere.

Background: Tumor invasion risk (TIR) is an important prognostic factor in nasopharyngeal carcinoma (NPC). We propose a novel prognostic analytic method for NPC based on a voxelwise analysis of TIR in a coordinate system of the nasopharynx. Methods: A stable nasopharynx coordinate system was constructed based on anatomical landmarks to obtain an accurate TIR profile for NPC. The coordinate system was validated by image registration of the lateral pterygoid muscle (LPM). The tumors were registered to the coordinate system through shift, scale, and rotation transformations. The voxelwise TIR map for NPC was obtained by superposition of all registered and mirrored tumor regions of interest. The minimum risk (MinR) point of the tumor region was used as an independent prognostic factor for NPC. The cutoff value was calculated with density plot and validated with restricted cubic splines (RCSs), and then the patients were divided into 2 groups for overall survival (OS) analysis. Results: The first voxelwise TIR map of NPC was obtained based on 778 patients. The OS of patients with a low TIR was 76.8% and was 92.6% for patients with a high TIR (P<0.001; hazard ratio (HR) =1/0.45; 95% CI: 0.27-0.77; adjusted P=0.004). Thus, patients with a low TIR had a poor prognosis, whereas patients with a high TIR had a good prognosis. The MinR may be better at grading the prognosis of patients compared to the American Joint Committee on Cancer (AJCC) staging or tumor/node (T/N) classification systems. Conclusions: The voxelwise TIR map provides a new method for the prognostic analysis of NPC. Potential clinical applications of voxelwise TIR mapping are clinical target volume (CTV) delineation and dose-painting for NPC.

: In order to explore a prognostic analysis method of postoperative tamoxifen treatment for breast cancer from mammography, the squeeze-and-convolutional Neural Network (SE-CNN) method was used to investigate the model of mammographic density automatic extraction from mammography and the prognostic effect of mammographic density on tamoxifen treatment for breast cancer. The results show that the mammographic density change rate of the subjects before and 15 months after surgery was extracted, and the mammographic density change rate cut value was obtained by density map method, and the subjects were divided into groups. The progression-free survival was HR: 2.654(95%CI,1.102-6.395), P =0.030. Patients with high mammographic density change rate had a better prognosis, while those with low mammographic density change rate had a worse prognosis. It is concluded that mammographic density change rate value can be a potential prognostic factor of postoperative tamoxifen treatment for breast cancer.

Background and Objective: An anatomical landmark is biologically meaningful point in medical images and often used for medical image registration. The purpose of this study is to automatically locate anatomical landmarks from 3D medical images. Methods: A two-step automatic location scheme of anatomical landmarks in 3D medical image was designed in this study. In the first step, the full convolutional neural network was used for slice detection from a 3D medical image. In the second step, the scale attention hourglass network was used for landmark location in the detected slice and could overcome the difficulty of similar anatomical structures and different image parameters. This method was implemented and tested on four stable anatomical landmarks in 3D head MRI. Results: A total of 500 and 300 3D head volumes were used for training and testing, respectively. Results showed that the slice detection accuracy reached 85.7% and that the maximum location error was less than one slice. The average accuracy of the four anatomical landmarks in the detected slice reached 87.2%, and the spatial distance was 2.4 +/- 2.4, which obtained better performance compared with hourglass network and feature pyramid networks. Conclusions: This method can be useful for locating anatomical landmarks in 3D head MRI and provides technical support for medical image registration and big data analysis. (C) 2021 Elsevier B.V. All rights reserved.

Do nasopharyngeal carcinoma (NPC) patients benefit from induction chemotherapy (IC)? This problem is of great clinical interest; however, it is difficult to obtain an accurate and interpretable model to inform IC decisions for NPC patients. In this study, a time-to-event supervised genetic algorithm was developed to obtain an IC decision-making model for NPC patients. In this algorithm, the fitness function is directly related to the time-to-event, which reflects the IC therapeutic effect for NPC. Then, the optimal models are obtained by stability and validation analysis. The comprehensive clinical model is determined by comprehensive feature analysis using the “or” operation. The overall survival for non-IC vs. IC patients in the potential benefit group was 63.4% vs. 81.5%, with p = 0.020, and the comprehensive clinical model exhibited good generalization ability. However, the benefits of OS according to the current NCCN guidelines are limited (p > 0.05). None of the possible processes of LASSO we tried could obtain the significant models validated in the testing cohort. The proposed method provides an interpretable model construction process, reasonable data grouping strategy, concise experimental design, and convenient clinical application. Moreover, we will develop a toolkit for the treatment decision-making model research to facilitate the use of clinicians and provide technical support for precision medicine.