Automated development of artificial neural networks for clinical purposes: Application for predicting the outcome of choledocholithiasis surgery

Conventional and machine learning-based analysis of age, body weight and body height significance in knot position-related thyrohyoid and cervical spine fractures in suicidal hangings

The thyrohyoid complex and cervical spine fracture distribution patterns may reflect the knot position as the force distribution by the noose to different neck regions may vary depending on it. Recently, machine learning models (MLm) were used to classify knot position through these fractures. The contribution of aging on the fracture susceptibility is better demonstrated, but data on body weight (BW) and height (BH) significance on this is more doubtful and MLm did not consider them. A retrospectively obtained autopsy data on sex, age, BW, BH and distribution of greater hyoid bone horn (GHH), superior thyroid cartilage horn (STH), and cervical spine fractures in 368 suicidal hangings were analyzed by standard statistics to determine association of the anthropometrics (age, BW, and BH) with the fracture occurrence, and by machine learning algorithms to determine if body weight and height improved MLm classification of hanging cases with typical and atypical knot positions. In the sample, unilateral GHH fracture was significantly more common in atypical hangings, while isolated STH fractures were more common in typical hangings. Age was a predictor of GHH fractures and BW of STH fractures, but BW poorly correlated with their number. BH was not a predictor of any thyrohyoid fracture. On the ROC curve analysis, the MLm that considered BW and BH did not perform statistically better than MLm that did not consider them. The study indicates that body weight and height are of no detrimental value in assessing the thyrohyoid and cervical spine fracture patterns in suicidal hangings.

A deep learning-driven method for safe and effective ERCP cannulation

PURPOSE

In recent years, the detection of the duodenal papilla and surgical cannula has become a critical task in computer-assisted endoscopic retrograde cholangiopancreatography (ERCP) cannulation operations. The complex surgical anatomy, coupled with the small size of the duodenal papillary orifice and its high similarity to the background, poses significant challenges to effective computer-assisted cannulation. To address these challenges, we present a deep learning-driven graphical user interface (GUI) to assist ERCP cannulation.

METHODS

Considering the characteristics of the ERCP scenario, we propose a deep learning method for duodenal papilla and surgical cannula detection, utilizing four swin transformer decoupled heads (4STDH). Four different prediction heads are employed to detect objects of different sizes. Subsequently, we integrate the swin transformer module to identify attention regions to explore prediction potential deeply. Moreover, we decouple the classification and regression networks, significantly improving the model's accuracy and robustness through the separation prediction. Simultaneously, we introduce a dataset on papilla and cannula (DPAC), consisting of 1840 annotated endoscopic images, which will be publicly available. We integrated 4STDH and several state-of-the-art methods into the GUI and compared them.

RESULTS

On the DPAC dataset, 4STDH outperforms state-of-the-art methods with an mAP of 93.2% and superior generalization performance. Additionally, the GUI provides real-time positions of the papilla and cannula, along with the planar distance and direction required for the cannula to reach the cannulation position.

CONCLUSION

We validate the GUI's performance in human gastrointestinal endoscopic videos, showing deep learning's potential to enhance the safety and efficiency of clinical ERCP cannulation.

Non-Invasive Prediction of Choledocholithiasis Using 1D Convolutional Neural Networks and Clinical Data

This paper introduces a novel one-dimensional convolutional neural network that utilizes clinical data to accurately detect choledocholithiasis, where gallstones obstruct the common bile duct. Swift and precise detection of this condition is critical to preventing severe complications, such as biliary colic, jaundice, and pancreatitis. This cutting-edge model was rigorously compared with other machine learning methods commonly used in similar problems, such as logistic regression, linear discriminant analysis, and a state-of-the-art random forest, using a dataset derived from endoscopic retrograde cholangiopancreatography scans performed at Olive View-University of California, Los Angeles Medical Center. The one-dimensional convolutional neural network model demonstrated exceptional performance, achieving 90.77% accuracy and 92.86% specificity, with an area under the curve of 0.9270. While the paper acknowledges potential areas for improvement, it emphasizes the effectiveness of the one-dimensional convolutional neural network architecture. The results suggest that this one-dimensional convolutional neural network approach could serve as a plausible alternative to endoscopic retrograde cholangiopancreatography, considering its disadvantages, such as the need for specialized equipment and skilled personnel and the risk of postoperative complications. The potential of the one-dimensional convolutional neural network model to significantly advance the clinical diagnosis of this gallstone-related condition is notable, offering a less invasive, potentially safer, and more accessible alternative.

Assessing the knot in a noose position by thyrohyoid and cervical spine fracture patterns in suicidal hangings using machine learning algorithms: A new insight into old dilemmas

Hanging is one of the most common suicide methods worldwide. Neck injuries that occur upon such neck compression - fractures of the thyrohyoid complex and cervical spine, occupy forensic pathologists for a long time. However, research failed to identify particular patterns of these injuries corresponding to the force distribution a ligature applies to the neck: the issue of reconstructing the knot in a noose position persists. So far, machine learning (ML) models were not utilized to classify knot positions and reconstruct this event. We conducted a single-institutional, retrospective study on 1235 autopsy cases of suicidal hanging, developed several ML models, and assessed their classification performance in a stepwise manner to discriminate between: 1. typical ('posterior) and atypical ('anterior' and 'lateral') hangings, 2. anterior and lateral hangings, and 3. left and right lateral hangings. The variable coding was based on the presence/absence of fractures of greater hyoid bone horns (GHH), superior thyroid cartilage horns (STH), and cervical spine. Subject age was considered. The models' parameters were optimized by the Genetic Algorithm. The accuracy of ML models in the first step was very modest (c. 60%) but increased subsequently: Multilayer Perceptron - Artificial Neural Network and k-Nearest Neighbors performed excellently discriminating between left and right lateral hangings (accuracy 91.8% and 90.6%, respectively). The latter is of great importance for clarifying probable hanging fracture biomechanics. Alongside the conventional inferential statistical analysis we performed, our results further indicate the association of the knot position with ipsilateral GHH and contralateral STH fractures in lateral hangings. Moreover, odds for unilateral GHH fracture, simultaneous GHH and STH fractures, and cervical spine fracture were significantly higher in atypical ('anterior' and 'lateral') hangings, compared to typical ('posterior') hangings.

WSES project on decision support systems based on artificial neural networks in emergency surgery

The article is a scoping review of the literature on the use of decision support systems based on artificial neural networks in emergency surgery. The authors present modern literature data on the effectiveness of artificial neural networks for predicting, diagnosing and treating abdominal emergency conditions: acute appendicitis, acute pancreatitis, acute cholecystitis, perforated gastric or duodenal ulcer, acute intestinal obstruction, and strangulated hernia. The intelligent systems developed at present allow a surgeon in an emergency setting, not only to check his own diagnostic and prognostic assumptions, but also to use artificial intelligence in complex urgent clinical cases. The authors summarize the main limitations for the implementation of artificial neural networks in surgery and medicine in general. These limitations are the lack of transparency in the decision-making process; insufficient quality educational medical data; lack of qualified personnel; high cost of projects; and the complexity of secure storage of medical information data. The development and implementation of decision support systems based on artificial neural networks is a promising direction for improving the forecasting, diagnosis and treatment of emergency surgical diseases and their complications.

Predicting sun protection measures against skin diseases using machine learning approaches

BACKGROUND

The substantial growth rate of skin cancer has necessitated adequate protection from solar radiation. Consequently, analyzing sun protection practices is an imperative research area in dermatology and pharmacology.

AIMS

This paper aims to analyze public sun-protection manners in the Arabian Peninsula regions.

METHODS

A simple random survey was conducted to assess public sun protection manners. Artificial neural network (ANN) and support vector machine (SVM) were selected from several machine learning algorithms to create the models for predicting public sun protection measures based on the prediction accuracy. Model performances were evaluated based on several performance indicators depending on the confusion matrices and receiver operating characteristic curves.

RESULTS

51% of the respondents have a low level, and 49% have a high level of sun protection practices. The results showed that the SVM performed considerably amended than the ANN for predicting the response. The relative importance of the predictors for the best predictive SVM model was also analyzed. The predictors are ranked as: the number of times having sunburnt >gender > use seat belt while driving/riding a vehicle >considers the UV index for personal sun exposure >income based on the expenses >sports/exercise activities >consciousness of the chance for having sunburnt on extended exposure to the sun >age > having any skin problem >nationality > skin type.

CONCLUSION

These identified significant predictors might be considered for developing an effective policy to increase public consciousness using proper protection from solar radiation's detrimental effect to rule out skin diseases.

Can artificial intelligence distinguish between malignant and benign mediastinal lymph nodes using sonographic features on EBUS images?

AIMS

This study aimed to develop a new intelligent diagnostic approach using an artificial neural network (ANN). Moreover, we investigated whether the learning-method-guided quantitative analysis approach adequately described mediastinal lymphadenopathies on endobronchial ultrasound (EBUS) images.

METHODS

In total, 345 lymph nodes (LNs) from 345 EBUS images were used as source input datasets for the application group. The group consisted of 300 and 45 textural patterns as input and output variables, respectively. The input and output datasets were processed using MATLAB. All these datasets were utilized for the training and testing of the ANN.

RESULTS

The best diagnostic accuracy was 82% of that obtained from the textural patterns of the LNs pattern (89% sensitivity, 72% specificity, and 78.2% area under the curve). The negative predictive values were 81% compared to the corresponding positive predictive values of 83%. Due to the application group's pattern-based evaluation, the LN pattern was statistically significant (p = .002).

CONCLUSIONS

The proposed intelligent approach could be useful in making diagnoses. Further development is required to improve the diagnostic accuracy of the visual interpretation.

Using machine learning methods to determine a typology of patients with HIV-HCV infection to be treated with antivirals

Several European countries have established criteria for prioritising initiation of treatment in patients infected with the hepatitis C virus (HCV) by grouping patients according to clinical characteristics. Based on neural network techniques, our objective was to identify those factors for HIV/HCV co-infected patients (to which clinicians have given careful consideration before treatment uptake) that have not being included among the prioritisation criteria. This study was based on the Spanish HERACLES cohort (NCT02511496) (April-September 2015, 2940 patients) and involved application of different neural network models with different basis functions (product-unit, sigmoid unit and radial basis function neural networks) for automatic classification of patients for treatment. An evolutionary algorithm was used to determine the architecture and estimate the coefficients of the model. This machine learning methodology found that radial basis neural networks provided a very simple model in terms of the number of patient characteristics to be considered by the classifier (in this case, six), returning a good overall classification accuracy of 0.767 and a minimum sensitivity (for the classification of the minority class, untreated patients) of 0.550. Finally, the area under the ROC curve was 0.802, which proved to be exceptional. The parsimony of the model makes it especially attractive, using just eight connections. The independent variable “recent PWID” is compulsory due to its importance. The simplicity of the model means that it is possible to analyse the relationship between patient characteristics and the probability of belonging to the treated group.

Evaluating Pedicle-Screw Instrumentation Using Decision-Tree Analysis Based on Pullout Strength

Study Design

A biomechanical study of pedicle-screw pullout strength.

Purpose

To develop a decision tree based on pullout strength for evaluating pedicle-screw instrumentation.

Overview of Literature

Clinically, a surgeon’s understanding of the holding power of a pedicle screw is based on perioperative intuition (which is like insertion torque) while inserting the screw. This is a subjective feeling that depends on the skill and experience of the surgeon. With the advent of robotic surgery, there is an urgent need for the creation of a patient-specific surgical planning system. A learning-based predictive model is needed to understand the sensitivity of pedicle-screw holding power to various factors.

Methods

Pullout studies were carried out on rigid polyurethane foam, representing extremely osteoporotic to normal bone for different insertion depths and angles of a pedicle screw. The results of these experimental studies were used to build a pullout-strength predictor and a decision tree using a machine-learning approach.

Results

Based on analysis of variance, it was found that all the factors under study had a significant effect (p <0.05) on the holding power of a pedicle screw. Of the various machine-learning techniques, the random forest regression model performed well in predicting the pullout strength and in creating a decision tree. Performance was evaluated, and a correlation coefficient of 0.99 was obtained between the observed and predicted values. The mean and standard deviation of the normalized predicted pullout strength for the confirmation experiment using the current model was 1.01±0.04.

Conclusions

The random forest regression model was used to build a pullout-strength predictor and decision tree. The model was able to predict the holding power of a pedicle screw for any combination of density, insertion depth, and insertion angle for the chosen range. The decision-tree model can be applied in patient-specific surgical planning and a decision-support system for spine-fusion surgery.

Assessment of cortical bone fracture resistance curves by fusing artificial neural networks and linear regression

Bone injures (BI) represents one of the major health problems, together with cancer and cardiovascular diseases. Assessment of the risks associated with BI is nontrivial since fragility of human cortical bone is varying with age. Due to restrictions for performing experiments on humans, only a limited number of fracture resistance curves (R-curves) for particular ages have been reported in the literature. This study proposes a novel decision support system for the assessment of bone fracture resistance by fusing various artificial intelligence algorithms. The aim was to estimate the R-curve slope, toughness threshold and stress intensity factor using the two input parameters commonly available during a routine clinical examination: patients age and crack length. Using the data from the literature, the evolutionary assembled Artificial Neural Network was developed and used for the derivation of Linear regression (LR) models of R-curves for arbitrary age. Finally, by using the patient (age)-specific LR models and diagnosed crack size one could estimate the risk of bone fracture under given physiological conditions. Compared to the literature, we demonstrated improved performances for estimating nonlinear changes of R-curve slope (R² = 0.82 vs. R² = 0.76) and Toughness threshold with ageing (R² = 0.73 vs. R² = 0.66).