Using deep transfer learning to detect scoliosis and spondylolisthesis from x-ray images

A machine learning based quantification system for automated diagnosis of lumbar spondylolisthesis on spinal X-rays

The automated diagnosis of lumbar spondylolisthesis lacks standardized criteria and the diagnostic of lumbar spondylolisthesis itself inherently relies on the subjective judgment of experts, resulting in a lack of standardized criteria. The objective of this study is to develop a novel, fully automated diagnostic system for lumbar spondylolisthesis. A two-stage system was developed, consisting of a Mask R-CNN with Prime Sample Attention (PISA) for vertebral segmentation in the first stage and a Light Gradient Boosting Machine (LGBM) for spondylolisthesis diagnosis in the second stage. The training data was developed by a total of 936 X-ray images including neutral, extension, and flexion lateral radiographs retrospectively gathered from 312 patients diagnosed with lumbar spondylolisthesis between January 2021 and March 2022. From a model perspective, there were a total of 4680 vertebrae, of which 1056 (22.6 %) were spondylolisthesis and the rest were normal. The Bbox mAP50, Bbox mAP75, Segm mAP50, and Segm mAP75 of Mask R-CNN with PISA were 0.9852, 0.9179, 0.9741, and 0.8957, respectively. The Accuracy, AUC, Recall, Precision, and F1-score of LGBM were 0.9660, 0.9843, 0.9020, 0.9020, and 0.9020, respectively. This study presents a robust system for the diagnosis of lumbar spondylolisthesis, providing accurate detection, classification, and localization of lumbar spondylolisthesis.

Heatmap-Based Active Shape Model for Landmark Detection in Lumbar X-ray Images

Medical staff inspect lumbar X-ray images to diagnose lumbar spine diseases, and the analysis process is currently automated using deep-learning techniques. The detection of landmarks is necessary in the automatic process of localizing the position and identifying the morphological features of the vertebrae. However, detection errors may occur owing to the noise and ambiguity of images, as well as individual variations in the shape of the lumbar vertebrae. This study proposes a method to improve the robustness of landmark detection results. This method assumes that landmarks are detected by a convolutional neural network-based two-step model consisting of Pose-Net and M-Net. The model generates a heatmap response to indicate the probable landmark positions. The proposed method then corrects the landmark positions using the heatmap response and active shape model, which employs statistical information on the landmark distribution. Experiments were conducted using 3600 lumbar X-ray images, and the results showed that the landmark detection error was reduced by the proposed method. The average value of maximum errors decreased by 5.58% after applying the proposed method, which combines the outstanding image analysis capabilities of deep learning with statistical shape constraints on landmark distribution. The proposed method could also be easily integrated with other techniques to increase the robustness of landmark detection results such as CoordConv layers and non-directional part affinity field. This resulted in a further enhancement in the landmark detection performance. These advantages can improve the reliability of automatic systems used to inspect lumbar X-ray images. This will benefit both patients and medical staff by reducing medical expenses and increasing diagnostic efficiency.

Explainable Deep-Learning-Based Gait Analysis of Hip-Knee Cyclogram for the Prediction of Adolescent Idiopathic Scoliosis Progression

Accurate prediction of scoliotic curve progression is crucial for guiding treatment decisions in adolescent idiopathic scoliosis (AIS). Traditional methods of assessing the likelihood of AIS progression are limited by variability and rely on static measurements. This study developed and validated machine learning models for classifying progressive and non-progressive scoliotic curves based on gait analysis using wearable inertial sensors. Gait data from 38 AIS patients were collected using seven inertial measurement unit (IMU) sensors, and hip-knee (HK) cyclograms representing inter-joint coordination were generated. Various machine learning algorithms, including support vector machine (SVM), random forest (RF), and novel deep convolutional neural network (DCNN) models utilizing multi-plane HK cyclograms, were developed and evaluated using 10-fold cross-validation. The DCNN model incorporating multi-plane HK cyclograms and clinical factors achieved an accuracy of 92% in predicting curve progression, outperforming SVM (55% accuracy) and RF (52% accuracy) models using handcrafted gait features. Gradient-based class activation mapping revealed that the DCNN model focused on the swing phase of the gait cycle to make predictions. This study demonstrates the potential of deep learning techniques, and DCNNs in particular, in accurately classifying scoliotic curve progression using gait data from wearable IMU sensors.

Detection and Localization of Spine Disorders from Plain Radiography

Spine disorders can cause severe functional limitations, including back pain, decreased pulmonary function, and increased mortality risk. Plain radiography is the first-line imaging modality to diagnose suspected spine disorders. Nevertheless, radiographical appearance is not always sufficient due to highly variable patient and imaging parameters, which can lead to misdiagnosis or delayed diagnosis. Employing an accurate automated detection model can alleviate the workload of clinical experts, thereby reducing human errors, facilitating earlier detection, and improving diagnostic accuracy. To this end, deep learning-based computer-aided diagnosis (CAD) tools have significantly outperformed the accuracy of traditional CAD software. Motivated by these observations, we proposed a deep learning-based approach for end-to-end detection and localization of spine disorders from plain radiographs. In doing so, we took the first steps in employing state-of-the-art transformer networks to differentiate images of multiple spine disorders from healthy counterparts and localize the identified disorders, focusing on vertebral compression fractures (VCF) and spondylolisthesis due to their high prevalence and potential severity. The VCF dataset comprised 337 images, with VCFs collected from 138 subjects and 624 normal images collected from 337 subjects. The spondylolisthesis dataset comprised 413 images, with spondylolisthesis collected from 336 subjects and 782 normal images collected from 413 subjects. Transformer-based models exhibited 0.97 Area Under the Receiver Operating Characteristic Curve (AUC) in VCF detection and 0.95 AUC in spondylolisthesis detection. Further, transformers demonstrated significant performance improvements against existing end-to-end approaches by 4-14% AUC (p-values < 10-13) for VCF detection and by 14-20% AUC (p-values < 10-9) for spondylolisthesis detection.

Quantitative evaluation model of variable diagnosis for chest X-ray images using deep learning

The purpose of this study is to demonstrate the use of a deep learning model in quantitatively evaluating clinical findings typically subject to uncertain evaluations by physicians, using binary test results based on routine protocols. A chest X-ray is the most commonly used diagnostic tool for the detection of a wide range of diseases and is generally performed as a part of regular medical checkups. However, when it comes to findings that can be classified as within the normal range but are not considered disease-related, the thresholds of physicians' findings can vary to some extent, therefore it is necessary to define a new evaluation method and quantify it. The implementation of such methods is difficult and expensive in terms of time and labor. In this study, a total of 83,005 chest X-ray images were used to diagnose the common findings of pleural thickening and scoliosis. A novel method for quantitatively evaluating the probability that a physician would judge the images to have these findings was established. The proposed method successfully quantified the variation in physicians' findings using a deep learning model trained only on binary annotation data. It was also demonstrated that the developed method could be applied to both transfer learning using convolutional neural networks for general image analysis and a newly learned deep learning model based on vector quantization variational autoencoders with high correlations ranging from 0.89 to 0.97.

Current situation and factors influencing physical fitness among adolescents aged 12 ∼ 15 in Shandong Province, China: A cross-sectional study

Adolescent physical fitness serves not merely as a current barometer of well-being but as a significant prognosticator of future health trajectories. Amidst the tumult of socioeconomic metamorphoses and pronounced lifestyle transitions enveloping China-mirroring global trends-the imperative to elucidate the present landscape of adolescent physical fitness intensifies. Moreover, discerning the myriad determinants underpinning it becomes paramount. In this context, our research endeavored to meticulously delineate the physical fitness milieu of adolescents residing in Shandong Province, systematically unpacking the multifarious influencers thereof. The insights garnered herein furnish an empirical foundation, primed to guide the sculpting of calibrated interventions, targeting the enhancement of health in this pivotal population cohort. In an extensive evaluative survey conducted in 2023 concerning the physical fitness of Shandong's student populace, a cohort of 33,211 adolescents aged 12 to 15 years was delineated utilizing a stratified random cluster sampling technique. This exercise meticulously quantified the physical fitness indices across diverse gender, age, and household registration classifications, subsequently computing the concomitant qualified rates. Employing multivariable logistic regression analysis, this investigation delved into the determinants modulating the adolescents' physical fitness qualified rate. For 2023, the aggregate fitness qualified rate stood at 91.94 % for the adolescents aged 12 ∼ 15 in Shandong Province. Gender-wise, female adolescents registered a qualified rate of 92.25 %, marginally eclipsing their male peers at 91.63 % (P < 0.05). An age-related trend in qualified rates was discernible, with marked variations across different age bands (P < 0.05): 91.37 % for 12-year-olds, 91.79 % for 13-year-olds, 91.81 % for 14-year-olds, and a zenith of 92.87 % for 15-year-olds. A geographical dichotomy emerged wherein rural adolescents distinctly outperformed their urban counterparts, notching up a 92.28 % qualified rate versus 91.64 % in urban settings (P < 0.05). The multivariable logistic regression analysis showed that after adjusting for gender, age, and household registration characteristics, adolescents had a lower odds of failing the physical fitness tests whose parents both liked physical exercises, whose parents supported children's participation in physical exercise, who participated in physical exercise sessions 3 ∼ 5 times per week or more than 5 times per week, who exercised for 0.5 ∼ 1 h each time or more than 1 h each time, who engaged in moderate intensity physical exercise, who slept 6 ∼ 8 h per day or more than 8 h per day, who consumed breakfast 3 ∼ 6 times per week or daily. On the other hand, adolescents had a higher odds of failing the physical fitness tests who always exposed to passive smoking, who spent 1 ∼ 3 h on screen per day or more than 3 h on screen per day, who spent more than 3 h doing homework per day, who consumed fast food 2 ∼ 3 times per week or more than 3 times per week. The physical fitness trajectory of adolescents residing within Shandong Province is tethered to a mosaic of determinants. This underscores the imperative for a synergistic strategy, harmonizing parental, scholastic, and societal vectors, to cultivate the salubrious maturation of this pivotal cohort.

Application of machine learning standardized integral area algorithm in measuring the scoliosis

This study was to develop a computer vision evaluation method to automatically measure the degree of scoliosis based on the machine learning algorithm. For the X-ray images of 204 patients with idiopathic scoliosis who underwent full-spine radiography, histogram equalization of original image was performed before a flipping method was used to magnify asymmetric elements, search for the global maximum pixel value in each line, and scan local maximal pixel value, with the intersection set of two point sets being regarded as candidate anchor points. All fine anchors were fitted with cubic spline algorithm to obtain the approximate curve of the spine, and the degree of scoliosis was measured by the standardized integral area. All measured data were analyzed. In manual measurement, the Cobb angle was 11.70-25.00 (20.15 ± 3.60), 25.20-44.70 (33.89 ± 5.41), and 45.10-49.40 (46.98 ± 1.25) in the mild, moderate and severe scoliosis group, respectively, whereas the value for the standardized integral area algorithm was 0.072-0.298 (0.185 ± 0.040), 0.100-0.399 (0.245 ± 0.050), and 0.246-0.901 (0.349 ± 0.181) in the mild, moderate and severe scoliosis group, respectively. Correlation analysis between the manual measurement of the Cobb angle and the evaluation of the standardized integral area algorithm demonstrated the Spearman correlation coefficient r = 0.643 (P < 0.001). There was a positive correlation between the manual measurement of the Cobb angle and the measurement of the standardized integral area value. Two methods had good consistency in evaluating the degree of scoliosis. ROC curve analysis of the standardized integral area algorithm to measure the degree of scoliosis showed he cutoff value of the standardized integral area algorithm was 0.20 for the moderate scoliosis with an AUC of 0.865, sensitivity 0.907, specificity 0.635, accuracy 0.779, positive prediction value 0.737 and negative prediction value 0.859, and the cutoff value of the standardized integral area algorithm was 0.40 for the severe scoliosis with an AUC of 0.873, sensitivity 0.188, specificity 1.00, accuracy 0.936, positive prediction value 1 and a negative prediction value 0.935. Using the standardized integral area as an independent variable and the Cobb angle as a dependent variable, a linear regression equation was established as Cobb angle = 13.36 + 70.54 × Standardized area, the model has statistical significance. In conclusion, the integrated area algorithm method of machine learning can quickly and efficiently assess the degree of scoliosis and is suitable for screening the degree of scoliosis in a large dataset as a useful supplement to the fine measurement of scoliosis Cobb angle.

Deep Transfer Learning with Enhanced Feature Fusion for Detection of Abnormalities in X-ray Images

Medical image classification poses significant challenges in real-world scenarios. One major obstacle is the scarcity of labelled training data, which hampers the performance of image-classification algorithms and generalisation. Gathering sufficient labelled data is often difficult and time-consuming in the medical domain, but deep learning (DL) has shown remarkable performance, although it typically requires a large amount of labelled data to achieve optimal results. Transfer learning (TL) has played a pivotal role in reducing the time, cost, and need for a large number of labelled images. This paper presents a novel TL approach that aims to overcome the limitations and disadvantages of TL that are characteristic of an ImageNet dataset, which belongs to a different domain. Our proposed TL approach involves training DL models on numerous medical images that are similar to the target dataset. These models were then fine-tuned using a small set of annotated medical images to leverage the knowledge gained from the pre-training phase. We specifically focused on medical X-ray imaging scenarios that involve the humerus and wrist from the musculoskeletal radiographs (MURA) dataset. Both of these tasks face significant challenges regarding accurate classification. The models trained with the proposed TL were used to extract features and were subsequently fused to train several machine learning (ML) classifiers. We combined these diverse features to represent various relevant characteristics in a comprehensive way. Through extensive evaluation, our proposed TL and feature-fusion approach using ML classifiers achieved remarkable results. For the classification of the humerus, we achieved an accuracy of 87.85%, an F1-score of 87.63%, and a Cohen's Kappa coefficient of 75.69%. For wrist classification, our approach achieved an accuracy of 85.58%, an F1-score of 82.70%, and a Cohen's Kappa coefficient of 70.46%. The results demonstrated that the models trained using our proposed TL approach outperformed those trained with ImageNet TL. We employed visualisation techniques to further validate these findings, including a gradient-based class activation heat map (Grad-CAM) and locally interpretable model-independent explanations (LIME). These visualisation tools provided additional evidence to support the superior accuracy of models trained with our proposed TL approach compared to those trained with ImageNet TL. Furthermore, our proposed TL approach exhibited greater robustness in various experiments compared to ImageNet TL. Importantly, the proposed TL approach and the feature-fusion technique are not limited to specific tasks. They can be applied to various medical image applications, thus extending their utility and potential impact. To demonstrate the concept of reusability, a computed tomography (CT) case was adopted. The results obtained from the proposed method showed improvements.

A dataset of COVID-19 x-ray chest images

The distinction between normal chest x-ray (CXR) images and abnormal ones containing features of disease (e.g., opacities, consolidation, etc.) is important for accurate medical diagnosis. CXR images contain valuable information concerning the physiological and pathological state of the lungs and airways. In addition, they provide information about the heart, chest bones, and some arteries (e.g., Aorta and pulmonary arteries). Deep learning artificial intelligence has taken great strides in the development of sophisticated medical models in a wide range of applications. More specifically, it has been shown to provide highly accurate diagnosis and detection tools. The dataset presented in this article contains the chest x-ray images from the examination of confirmed COVID-19 subjects, who were admitted for a multiday stay at a local hospital in northern Jordan. To provide a diverse dataset, only one CXR image per subject was included in the data. The dataset can be used for the development of automated methods that detect COVID-19 from CXR images (COVID-19 vs. normal) and distinguish pneumonia caused by COVID-19 from other pulmonary diseases. ©202x The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Classification Models of Action Research Arm Test Activities in Post-Stroke Patients Based on Human Hand Motion

The Action Research Arm Test (ARAT) presents a ceiling effect that prevents the detection of improvements produced with rehabilitation treatments in stroke patients with mild finger joint impairments. The aim of this study was to develop classification models to predict whether activities with similar ARAT scores were performed by a healthy subject or by a subject post-stroke using the extension and flexion angles of 11 finger joints as features. For this purpose, we used three algorithms: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN). The dataset presented class imbalance, and the classification models presented a low recall, especially in the stroke class. Therefore, we implemented class balance using Borderline-SMOTE. After data balancing the classification models showed significantly higher accuracy, recall, f1-score, and AUC. However, after data balancing, the SVM classifier showed a higher performance with a precision of 98%, a recall of 97.5%, and an AUC of 0.996. The results showed that classification models based on human hand motion features in combination with the oversampling algorithm Borderline-SMOTE achieve higher performance. Furthermore, our study suggests that there are differences in ARAT activities performed between healthy and post-stroke individuals that are not detected by the ARAT scoring process.

Classification of Corn Diseases from Leaf Images Using Deep Transfer Learning

Corn is a mass-produced agricultural product that plays a major role in the food chain and many agricultural products in addition to biofuels. Furthermore, households in poor countries may depend on small-scale corn cultivation for their basic needs. However, corn crops are vulnerable to diseases, which greatly affects farming yields. Moreover, extreme weather conditions and unseasonable temperatures can accelerate the spread of diseases. The pervasiveness and ubiquity of technology have allowed for the deployment of technological innovations in many areas. Particularly, applications powered by artificial intelligence algorithms have established themselves in many disciplines relating to image, signal, and sound recognition. In this work, we target the application of deep transfer learning in the classification of three corn diseases (i.e., Cercospora leaf spot, common rust, and northern leaf blight) in addition to the healthy plants. Using corn leaf image as input and convolutional neural networks models, no preprocessing or explicit feature extraction was required. Transfer learning using well-established and well-designed deep learning models was performed and extensively evaluated using multiple scenarios for splitting the data. In addition, the experiments were repeated 10 times to account for variability in picking random choices. The four classes were discerned with a mean accuracy of 98.6%. This and the other performance metrics exhibit values that make it feasible to build and deploy applications that can aid farmers and plant pathologists to promptly and accurately perform disease identification and apply the correct remedies.

Detection of developmental dysplasia of the hip in X-ray images using deep transfer learning

Background

Developmental dysplasia of the hip (DDH) is a relatively common disorder in newborns, with a reported prevalence of 1–5 per 1000 births. It can lead to developmental abnormalities in terms of mechanical difficulties and a displacement of the joint (i.e., subluxation or dysplasia). An early diagnosis in the first few months from birth can drastically improve healing, render surgical intervention unnecessary and reduce bracing time. A pelvic X-ray inspection represents the gold standard for DDH diagnosis. Recent advances in deep learning artificial intelligence have enabled the use of many image-based medical decision-making applications. The present study employs deep transfer learning in detecting DDH in pelvic X-ray images without the need for explicit measurements.

Methods

Pelvic anteroposterior X-ray images from 354 subjects (120 DDH and 234 normal) were collected locally at two hospitals in northern Jordan. A system that accepts these images as input and classifies them as DDH or normal was developed using thirteen deep transfer learning models. Various performance metrics were evaluated in addition to the overfitting/underfitting behavior and the training times.

Results

The highest mean DDH detection accuracy was 96.3% achieved using the DarkNet53 model, although other models achieved comparable results. A common theme across all the models was the extremely high sensitivity (i.e., recall) value at the expense of specificity. The F1 score, precision, recall and specificity for DarkNet53 were 95%, 90.6%, 100% and 94.3%, respectively.

Conclusions

Our automated method appears to be a highly accurate DDH screening and diagnosis method. Moreover, the performance evaluation shows that it is possible to further improve the system by expanding the dataset to include more X-ray images.

On the Automatic Detection and Classification of Skin Cancer Using Deep Transfer Learning

Skin cancer (melanoma and non-melanoma) is one of the most common cancer types and leads to hundreds of thousands of yearly deaths worldwide. It manifests itself through abnormal growth of skin cells. Early diagnosis drastically increases the chances of recovery. Moreover, it may render surgical, radiographic, or chemical therapies unnecessary or lessen their overall usage. Thus, healthcare costs can be reduced. The process of diagnosing skin cancer starts with dermoscopy, which inspects the general shape, size, and color characteristics of skin lesions, and suspected lesions undergo further sampling and lab tests for confirmation. Image-based diagnosis has undergone great advances recently due to the rise of deep learning artificial intelligence. The work in this paper examines the applicability of raw deep transfer learning in classifying images of skin lesions into seven possible categories. Using the HAM1000 dataset of dermoscopy images, a system that accepts these images as input without explicit feature extraction or preprocessing was developed using 13 deep transfer learning models. Extensive evaluation revealed the advantages and shortcomings of such a method. Although some cancer types were correctly classified with high accuracy, the imbalance of the dataset, the small number of images in some categories, and the large number of classes reduced the best overall accuracy to 82.9%.