Keratoconus disease classification with multimodel fusion and vision transformer: a pretrained model approach

OBJECTIVE

Our objective is to develop a novel keratoconus image classification system that leverages multiple pretrained models and a transformer architecture to achieve state-of-the-art performance in detecting keratoconus.

METHODS AND ANALYSIS

Three pretrained models were used to extract features from the input images. These models have been trained on large datasets and have demonstrated strong performance in various computer vision tasks.The extracted features from the three pretrained models were fused using a feature fusion technique. This fusion aimed to combine the strengths of each model and capture a more comprehensive representation of the input images. The fused features were then used as input to a vision transformer, a powerful architecture that has shown excellent performance in image classification tasks. The vision transformer learnt to classify the input images as either indicative of keratoconus or not.The proposed method was applied to the Shahroud Cohort Eye collection and keratoconus detection dataset. The performance of the model was evaluated using standard evaluation metrics such as accuracy, precision, recall and F1 score.

RESULTS

The research results demonstrated that the proposed model achieved higher accuracy compared with using each model individually.

CONCLUSION

The findings of this study suggest that the proposed approach can significantly improve the accuracy of image classification models for keratoconus detection. This approach can serve as an effective decision support system alongside physicians, aiding in the diagnosis of keratoconus and potentially reducing the need for invasive procedures such as corneal transplantation in severe cases.

Development of glaucoma predictive model and risk factors assessment based on supervised models

Objectives

To develop and to propose a machine learning model for predicting glaucoma and identifying its risk factors.

Method

Data analysis pipeline is designed for this study based on Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology. The main steps of the pipeline include data sampling, preprocessing, classification and evaluation and validation. Data sampling for providing the training dataset was performed with balanced sampling based on over-sampling and under-sampling methods. Data preprocessing steps were missing value imputation and normalization. For classification step, several machine learning models were designed for predicting glaucoma including Decision Trees (DTs), K-Nearest Neighbors (K-NN), Support Vector Machines (SVM), Random Forests (RFs), Extra Trees (ETs) and Bagging Ensemble methods. Moreover, in the classification step, a novel stacking ensemble model is designed and proposed using the superior classifiers.

Results

The data were from Shahroud Eye Cohort Study including demographic and ophthalmology data for 5190 participants aged 40-64 living in Shahroud, northeast Iran. The main variables considered in this dataset were 67 demographics, ophthalmologic, optometric, perimetry, and biometry features for 4561 people, including 4474 non-glaucoma participants and 87 glaucoma patients. Experimental results show that DTs and RFs trained based on under-sampling of the training dataset have superior performance for predicting glaucoma than the compared single classifiers and bagging ensemble methods with the average accuracy of 87.61 and 88.87, the sensitivity of 73.80 and 72.35, specificity of 87.88 and 89.10 and area under the curve (AUC) of 91.04 and 94.53, respectively. The proposed stacking ensemble has an average accuracy of 83.56, a sensitivity of 82.21, a specificity of 81.32, and an AUC of 88.54.

Conclusions

In this study, a machine learning model is proposed and developed to predict glaucoma disease among persons aged 40-64. Top predictors in this study considered features for discriminating and predicting non-glaucoma persons from glaucoma patients include the number of the visual field detect on perimetry, vertical cup to disk ratio, white to white diameter, systolic blood pressure, pupil barycenter on Y coordinate, age, and axial length.

Proposing a machine-learning based method to predict stillbirth before and during delivery and ranking the features: nationwide retrospective cross-sectional study

Background

Stillbirth is defined as fetal loss in pregnancy beyond 28 weeks by WHO. In this study, a machine-learning based method is proposed to predict stillbirth from livebirth and discriminate stillbirth before and during delivery and rank the features.

Method

A two-step stack ensemble classifier is proposed for classifying the instances into stillbirth and livebirth at the first step and then, classifying stillbirth before delivery from stillbirth during the labor at the second step. The proposed SE has two consecutive layers including the same classifiers. The base classifiers in each layer are decision tree, Gradient boosting classifier, logistics regression, random forest and support vector machines which are trained independently and aggregated based on Vote boosting method. Moreover, a new feature ranking method is proposed in this study based on mean decrease accuracy, Gini Index and model coefficients to find high-ranked features.

Results

IMAN registry dataset is used in this study considering all births at or beyond 28th gestational week from 2016/04/01 to 2017/01/01 including 1,415,623 live birth and 5502 stillbirth cases. A combination of maternal demographic features, clinical history, fetal properties, delivery descriptors, environmental features, healthcare service provider descriptors and socio-demographic features are considered. The experimental results show that our proposed SE outperforms the compared classifiers with the average accuracy of 90%, sensitivity of 91%, specificity of 88%. The discrimination of the proposed SE is assessed and the average AUC of ±95%, CI of 90.51% ±1.08 and 90% ±1.12 is obtained on training dataset for model development and test dataset for external validation, respectively. The proposed SE is calibrated using isotopic nonparametric calibration method with the score of 0.07. The process is repeated 10,000 times and AUC of SE classifiers using random different training datasets as null distribution. The obtained p-value to assess the specificity of the proposed SE is 0.0126 which shows the significance of the proposed SE.

Conclusions

Gestational age and fetal height are two most important features for discriminating livebirth from stillbirth. Moreover, hospital, province, delivery main cause, perinatal abnormality, miscarriage number and maternal age are the most important features for classifying stillbirth before and during delivery.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-021-03658-z.

CNFE-SE: a novel approach combining complex network-based feature engineering and stacked ensemble to predict the success of intrauterine insemination and ranking the features

Background

Intrauterine Insemination (IUI) outcome prediction is a challenging issue which the assisted reproductive technology (ART) practitioners are dealing with. Predicting the success or failure of IUI based on the couples' features can assist the physicians to make the appropriate decision for suggesting IUI to the couples or not and/or continuing the treatment or not for them. Many previous studies have been focused on predicting the in vitro fertilization (IVF) and intracytoplasmic sperm injection (ICSI) outcome using machine learning algorithms. But, to the best of our knowledge, a few studies have been focused on predicting the outcome of IUI. The main aim of this study is to propose an automatic classification and feature scoring method to predict intrauterine insemination (IUI) outcome and ranking the most significant features.

Methods

For this purpose, a novel approach combining complex network-based feature engineering and stacked ensemble (CNFE-SE) is proposed. Three complex networks are extracted considering the patients' data similarities. The feature engineering step is performed on the complex networks. The original feature set and/or the features engineered are fed to the proposed stacked ensemble to classify and predict IUI outcome for couples per IUI treatment cycle. Our study is a retrospective study of a 5-year couples' data undergoing IUI. Data is collected from Reproductive Biomedicine Research Center, Royan Institute describing 11,255 IUI treatment cycles for 8,360 couples. Our dataset includes the couples' demographic characteristics, historical data about the patients' diseases, the clinical diagnosis, the treatment plans and the prescribed drugs during the cycles, semen quality, laboratory tests and the clinical pregnancy outcome.

Results

Experimental results show that the proposed method outperforms the compared methods with Area under receiver operating characteristics curve (AUC) of 0.84 ± 0.01, sensitivity of 0.79 ± 0.01, specificity of 0.91 ± 0.01, and accuracy of 0.85 ± 0.01 for the prediction of IUI outcome.

Conclusions

The most important predictors for predicting IUI outcome are semen parameters (sperm motility and concentration) as well as female body mass index (BMI).

Prediction of Cranial Radiotherapy Treatment in Pediatric Acute Lymphoblastic Leukemia Patients Using Machine Learning: A Case Study at MAHAK Hospital

Background:

Acute Lymphoblastic Leukemia (ALL) is the most common blood disease in children and is responsible for the most deaths amongst children. Due to major improvements in the treatment protocols in the 50-years period, the survivability of this disease has witnessed dramatic rise until this date which is about 90 percent. There are many investigations tending to indicate the efficiency of cranial radiotherapy found out that without that, outcome of the patients did not change and even it improved at some cases.

Methods:

the main aim of this study is predicting cranial radiotherapy treatment in pediatric acute lymphoblastic leukemia patients using machine learning. Scope of this paper is intertwined with predicting the necessity of one of the treatment modalities that has been used for many years for this group of patients named Cranial Radiotherapy (CRT). For this purpose, a case study is considered at Mahak charity hospital. In this paper, our focus is on ALL patients aged 0 to 17 treated at Mahak hospital, one of the best centers for treatment of childhood malignancies in Iran. Dataset analyzed in this study is gathered by the research team from patient’s paper-based files. Our dataset consists of 241 observations on patients with 31 attributes after the data cleaning process. Our designed machine learning model for predicting cranial radiotherapy treatment in pediatric acute lymphoblastic leukemia patients is a stacked ensemble classifier of independently strong models with a meta-learner to tune the weights and parameters of the base classifiers.

Results:

The stacked ensemble classifier show highly reasonable performance with AUC of 87.52%. Moreover, the attributes are ranked based on their predictive power and the most important variable for CRT necessity prediction is the disease relapse.

Conclusion:

In order to conclude, derived from previous studies regarding CRT it is not only cost-effective but also more healthy to eradicate the use of CRT for the treatment of childhood ALL. Furthermore, it is valuable to increase the clinical databases by creating more synthetic health databases not only for research purposes but also for physicians to keep track of their patient’s status.

Proposing a novel unsupervised stack ensemble of deep and conventional image segmentation (SEDCIS) method for localizing vitiligo lesions in skin images

BACKGROUND

Vitiligo is an acquired pigmentary skin disorder characterized by depigmented macules and patches which brings many challenges for the patients suffering from. For vitiligo severity assessment, several scoring methods have been proposed based on morphometry and colorimetry. But, all methods suffer from much inter- and intra-observer variations for estimating the depigmented area. For all mentioned assessment methods of vitiligo disorder, accurate segmentation of the skin images for lesion detection and localization is required. The image segmentation for localizing vitiligo skin lesions has many challenges because of illumination variation, different shapes and sizes of vitiligo lesions, vague lesion boundaries and skin hairs and vignette effects. The manual image segmentation is a tedious and time-consuming task. Therefore, using automatic image segmentation methods for lesion detection is necessarily required.

MATERIALS AND METHODS

In this study, a novel unsupervised stack ensemble of deep and conventional image segmentation (SEDCIS) methods is proposed for localizing vitiligo lesions in skin images. Unsupervised segmentation methods do not require prior manual segmentation of vitiligo lesions which is a tedious and time-consuming task with intra- and inter-observer variations.

RESULTS

Our collected dataset includes 877 images taken from 21 patients with the resolution of 5760*3840 pixels suffering from vitiligo disorder. Experimental results show that SEDCIS outperforms the compared methods with accuracy of 97%, sensitivity of 98%, specificity of 96%, area overlapping of 94%, and Dice index of 97%.

CONCLUSION

The proposed method can segment vitiligo lesions with highly reasonable performance and can be used for assessing the vitiligo lesion surface.

Studying the Effects of Systemic Inflammatory Markers and Drugs on AVF Longevity through a Novel Clinical Intelligent Framework

Although arteriovenous fistula is the preferred vascular access method, it has challenges in three phases of planning, maturation, and maintenance. We looked at the root of fistula challenges in the maintenance phase and found traces of inflammation. Accordingly, we investigated the role of systemic inflammation in this phase to understand its effects on post-maturation function and extract knowledge to help extend fistula longevity. Previous studies on longevity of fistula have focused entirely on statistical tests, and since they put limitations on data, we also used a data mining framework for data analysis. For prediction, we used Decision Tree, Random Forest, and Support Vector Machines, and for inferential analysis, we used Wilcoxon and Chi-squared tests. We analyzed the archived data of 119 hemodialysis patients. In these data, independent variables were serum inflammatory markers, serum metabolic values, anti-inflammatory drugs, and demographic characteristics, and the dependent variable was fistula longevity separated in classes of equal to or greater than four and less than four years. Both predictive and inferential approaches have shown that serum inflammatory markers had no significant involvement in fistula longevity, but some anti-inflammatory drugs were effective. The results have shown that blood tests and drug variables, alone or together, could predict longevity class by 100% accuracy. This prediction can help surgeons make better decisions in selecting patients for fistula creation. Also, the extracted knowledge can provide guidelines for post-maturation disorders.

Proposing feature engineering method based on deep learning and K-NNs for ECG beat classification and arrhythmia detection

Arrhythmia is slow, fast or irregular heartbeat. Manual ECG assessment and disease classification is an error-prone task because of vast differences in ECG morphology and difficulty in accurate identifying ECG components. Moreover, proposing a computer-aided diagnosis system for heartbeat classification can be useful when access to medical care centers is difficult or impossible. Therefore, the main aim of this study is classifying ECG beats for arrhythmia detection (four beat classes are considered). Previous studies have proposed different methods based on traditional machine learning and/or deep learning. In this paper, a novel feature engineering method is proposed based on deep learning and K-NNs. The features extracted by our proposed method are classified with different classifiers such as decision trees, SVMs with different kernels and random forests. Our proposed method has reasonably good performance for beat classification and achieves the average Accuracy of 99.77%, AUC of 99.99%, Precision of 99.75% and Recall of 99.30% using fivefold Cross Validation strategy. The main advantage of the proposed method is its low computational time compared to training deep learning models from scratch and its high accuracy compared to the traditional machine learning models. The strength and suitability of the proposed method for feature extraction is shown by the high balance between sensitivity and specificity.

Identification and weighting of kidney allocation criteria: a novel multi-expert fuzzy method

Background

Kidney allocation is a multi-criteria and complex decision-making problem, which should also consider ethical issues in addition to the medical aspects. Leading countries in this field use a point scoring system to allocate kidneys. Hence, the purpose of this study is to identify and weight the kidney allocation criteria considering the balance between utility and equity.

Methods

To do this, a new fuzzy hybrid approach is proposed, which consists of two steps: In the first step, Fuzzy Delphi Method (FDM) is used to identify the effective criteria in the kidney allocation algorithm. In the second step, Intuitionistic Fuzzy Analytic Hierarchy Process (IF-AHP) is employed to determine the weight of the criteria.

Results

The results showed that the highest weight belongs to “Medical emergency” criterion and the lowest weight to “5 HLA mismatches”, which is similar to Euro-transplant kidney allocation system (ETKAS). The developed method is evaluated in two steps. First, the proposed model is implemented using a real case study from the Iranian Kidney Allocation System. It was shown that the proposed model has the potential to improve allocation outcome. Second, the proposed model’s superiority to the current model is approved by the experts using the results display in the profile matrix. Finally, sensitivity analysis is performed to check the robustness of the proposed model.

Conclusions

This paper contributes to the kidney allocation literature by doing the following: (a) developing a comprehensive framework for identification and weightings of criteria for kidney allocation, (b) using, for the first time, the IF-AHP technique to consider hesitancy of decision makers and uncertainty in organ allocation, and (c) proposing an appropriate framework for the countries that intend to improve or modify their organ allocation system.

Electronic supplementary material

The online version of this article (10.1186/s12911-019-0892-y) contains supplementary material, which is available to authorized users.

An intelligent warning model for early prediction of cardiac arrest in sepsis patients

BACKGROUND

Sepsis-associated cardiac arrest is a common issue with the low survival rate. Early prediction of cardiac arrest can provide the time required for intervening and preventing its onset in order to reduce mortality. Several studies have been conducted to predict cardiac arrest using machine learning. However, no previous research has used machine learning for predicting cardiac arrest in adult sepsis patients. Moreover, the potential of some techniques, including ensemble algorithms, has not yet been addressed in improving the prediction outcomes. It is required to find methods for generating high-performance predictions with sufficient time lapse before the arrest. In this regard, various variables and parameters should also been examined.

OBJECTIVE

The aim was to use machine learning in order to propose a cardiac arrest prediction model for adult patients with sepsis. It is required to predict the arrest several hours before the incidence with high efficiency. The other goal was to investigate the effect of the time series dynamics of vital signs on the prediction of cardiac arrest.

METHOD

30 h clinical data of every sepsis patients were extracted from Mimic III database (79 cases, 4532 controls). Three datasets (multivariate, time series and combined) were created. Various machine learning models for six time groups were trained on these datasets. The models included classical techniques (SVM, decision tree, logistic regression, KNN, GaussianNB) and ensemble methods (gradient Boosting, XGBoost, random forest, balanced bagging classifier and stacking). Proper solutions were proposed to address the challenges of missing values, imbalanced classes of data and irregularity of time series.

RESULTS

The best results were obtained using a stacking algorithm and multivariate dataset (accuracy = 0.76, precision = 0.19, sensitivity = 0.77, f1-score = 0.31, AUC= 0.82). The proposed model predicts the arrest incidence of up to six hours earlier with the accuracy and sensitivity over 70%.

CONCLUSION

We illustrated that machine learning techniques, especially ensemble algorithms have high potentials to be used in prognostic systems for sepsis patients. The proposed model, in comparison with the exiting warning systems including APACHE II and MEWS, significantly improved the evaluation criteria. According to the results, the time series dynamics of vital signs are of great importance in the prediction of cardiac arrest incidence in sepsis patients.

A Novel Unsupervised Approach for Minimally-invasive Video Segmentation

Temporal segmentation of laparoscopic video is the first step toward identifying anomalies and interrupts, recognizing actions, annotating video and assessing the surgeons' learning curve. In this paper, a novel approach for temporal segmentation of minimally-invasive videos (MIVS) is proposed. Illumination variation, shadowing, dynamic backgrounds and tissue respiratory motion make it challenging to extract information from laparoscopic videos. These challenges if not properly addressed could increase the errors of data extraction modules. Therefore, in MIVS, several data sets are extracted from laparoscopic videos using different methods to alleviate error effects of data extraction modules on MIVS performance. Each extracted data set is segmented temporally with Genetic Algorithm (GA) after outlier removal. Three different cost functions are examined as objective function of GA. The correlation coefficient is calculated between objective values of the solutions visited by GA and their corresponding performance measures. Performance measures include detection rate, recognition rate and accuracy. Cost functions having negative correlations with all mentioned performance measures are selected. Finally, a multi-objective GA is executed on the data sets to optimize the selected cost functions. MIVS is tested on laparoscopic videos of varicocele and ureteropelvic junction obstruction surgeries collected from hasheminejad kidney center. Experimental results demonstrate that MIVS outperforms the state-of-the-art methods in terms of accuracy, detection rate and recognition rate.