Neural network approaches to grade adult depression

Proof of concept: Predicting distress in cancer patients using back propagation neural network (BPNN)

Background

Research findings suggest that a significant proportion of individuals diagnosed with cancer, ranging from 25% to 60%, experience distress and require access to psycho-oncological services. Until now, only contemporary approaches, such as logistic regression, have been used to determine predictors of distress in oncological patients. To improve individual prediction accuracy, novel approaches are required. We aimed to establish a prediction model for distress in cancer patients based on a back propagation neural network (BPNN).

Methods

Retrospective data was gathered from a cohort of 3063 oncological patients who received diagnoses and treatment spanning the years 2011-2019. The distress thermometer (DT) has been used as screening instrument. Potential predictors of distress were identified using logistic regression. Subsequently, a prediction model for distress was developed using BPNN.

Results

Logistic regression identified 13 significant independent variables as predictors of distress, including emotional, physical and practical problems. Through repetitive data simulation processes, it was determined that a 3-layer BPNN with 8 neurons in the hidden layer demonstrates the highest level of accuracy as a prediction model. This model exhibits a sensitivity of 79.0%, specificity of 71.8%, positive predictive value of 78.9%, negative predictive value of 71.9%, and an overall coincidence rate of 75.9%.

Conclusion

The final BPNN model serves as a compelling proof of concept for leveraging artificial intelligence in predicting distress and its associated risk factors in cancer patients. The final model exhibits a remarkable level of discrimination and feasibility, underscoring its potential for identifying patients vulnerable to distress.

Reading Wishes from the Lips: Cancer Patients' Need for Psycho-Oncological Support during Inpatient and Outpatient Treatment

Background: Psycho-oncological support (PO) is an effective measure to reduce distress and improve the quality of life in patients with cancer. Currently, there are only a few studies investigating the (expressed) wish for PO. The aim of this study was to evaluate the number of patients who request PO and to identify predictors for the wish for PO. Methods: Data from 3063 cancer patients who had been diagnosed and treated at a Comprehensive Cancer Center between 2011 and 2019 were analyzed retrospectively. Potential predictors for the wish for PO were identified using logistic regression. As a novelty, a Back Propagation Neural Network (BPNN) was applied to establish a prediction model for the wish for PO. Results: In total, 1752 patients (57.19%) had a distress score above the cut-off and 14.59% expressed the wish for PO. Patients’ requests for pastoral care (OR = 13.1) and social services support (OR = 5.4) were the strongest predictors of the wish for PO. Patients of the female sex or who had a current psychiatric diagnosis, opioid treatment and malignant neoplasms of the skin and the hematopoietic system also predicted the wish for PO, while malignant neoplasms of digestive organs and older age negatively predicted the wish for PO. These nine significant predictors were used as input variables for the BPNN model. BPNN computations indicated that a three-layer network with eight neurons in the hidden layer is the most precise prediction model. Discussion: Our results suggest that the identification of predictors for the wish for PO might foster PO referrals and help cancer patients reduce barriers to expressing their wish for PO. Furthermore, the final BPNN prediction model demonstrates a high level of discrimination and might be easily implemented in the hospital information system.

Multi-Model Adaptation Learning With Possibilistic Clustering Assumption for EEG-Based Emotion Recognition

In machine learning community, graph-based semi-supervised learning (GSSL) approaches have attracted more extensive research due to their elegant mathematical formulation and good performance. However, one of the reasons affecting the performance of the GSSL method is that the training data and test data need to be independently identically distributed (IID); any individual user may show a completely different encephalogram (EEG) data in the same situation. The EEG data may be non-IID. In addition, noise/outlier sensitiveness still exist in GSSL approaches. To these ends, we propose in this paper a novel clustering method based on structure risk minimization model, called multi-model adaptation learning with possibilistic clustering assumption for EEG-based emotion recognition (MA-PCA). It can effectively minimize the influence from the noise/outlier samples based on different EEG-based data distribution in some reproduced kernel Hilbert space. Our main ideas are as follows: (1) reducing the negative impact of noise/outlier patterns through fuzzy entropy regularization, (2) considering the training data and test data are IID and non-IID to obtain a better performance by multi-model adaptation learning, and (3) the algorithm implementation and convergence theorem are also given. A large number of experiments and deep analysis on real DEAP datasets and SEED datasets was carried out. The results show that the MA-PCA method has superior or comparable robustness and generalization performance to EEG-based emotion recognition.

Neural Network-Based Diagnostic Tool for Identifying the Factors Responsible for Depression

The paper aims at establishing the output-to-input relationship of the real-life adult depression data using a neural network (NN) model. The said model has been developed to diagnose and detect the associated severity (grade) of the illness. An intelligent NN-based reverse model has been trained through batch mode and put to test on another set of real-life data. Reverse mapping of this model has been developed to isolate significantly contributing input components (factors) for any given case to expedite the preventive procedure for further deterioration as well as the start of treatment.

A Novel Internet of Things Framework Integrated with Real Time Monitoring for Intelligent Healthcare Environment

During mammogram screening, there is a higher probability that detection of cancers is missed, and more than 16 percentage of breast cancer is not detected by radiologists. This problem can be solved by employing image processing algorithms which enhances the accuracy of the diagnostic through image segmentation which reduces the misclassified malignant cancers. By employing segmentation, the unnecessary regions in the breast close to the boundary between the breast tissue and segmented pectoral muscle can be removed, therefore enhancing the accuracy the calculation as well as feature estimation. In-order to enhance the accuracy of classification, the proposed classifier integrates the decision trees and neural network into a system to report the progress of the breast cancer patients in an appropriate manner with the help of technology used in healthcare system. The proposed classifier successfully demonstrated that it achieved more accurate prediction when compared with other widely used algorithms, namely, K-Nearest Neighbors, Support Vector Machine and Naive Bayes algorithm.

Plausibility of a Neural Network Classifier-Based Neuroprosthesis for Depression Detection via Laughter Records

The present work explores the diagnostic performance for depression of neural network classifiers analyzing the sound structures of laughter as registered from clinical patients and healthy controls. The main methodological novelty of this work is that simple sound variables of laughter are used as inputs, instead of electrophysiological signals or local field potentials (LFPs) or spoken language utterances, which are the usual protocols up-to-date. In the present study, involving 934 laughs from 30 patients and 20 controls, four different neural networks models were tested for sensitivity analysis, and were additionally trained for depression detection. Some elementary sound variables were extracted from the records: timing, fundamental frequency mean, first three formants, average power, and the Shannon-Wiener entropy. In the results obtained, two of the neural networks show a diagnostic discrimination capability of 93.02 and 91.15% respectively, while the third and fourth ones have an 87.96 and 82.40% percentage of success. Remarkably, entropy turns out to be a fundamental variable to distinguish between patients and controls, and this is a significant factor which becomes essential to understand the deep neurocognitive relationships between laughter and depression. In biomedical terms, our neural network classifier-based neuroprosthesis opens up the possibility of applying the same methodology to other mental-health and neuropsychiatric pathologies. Indeed, exploring the application of laughter in the early detection and prognosis of Alzheimer and Parkinson would represent an enticing possibility, both from the biomedical and the computational points of view.

HALF-WAVE SEGMENT FEATURE EXTRACTION OF EEG SIGNALS OF PATIENTS WITH DEPRESSION AND PERFORMANCE EVALUATION OF NEURAL NETWORK CLASSIFIERS

A detailed understanding of key signal characteristics has enabled the use of artificial neural networks (ANN) for feature detection and classification of EEG signals in clinical research. The present study is performed to classify EEG signals of normal and depression patients with wavelet parameters as key input features. The characteristics of depression cannot be made out by visual inspection of EEG records unlike epilepsy which is well characterized by sudden recurrent and transient waveforms. In this study, a comparison is made between the performance of feedforward neural network (FFNN) and probabilistic neural network (PNN) while classifying the EEG signals of normal and depression patients. Classification capabilities of both the methods are validated with the EEG recordings from 30 normal controls and 30 depression patients. One-way ANOVA provided a statistical significant difference between the two classes of EEG signals recorded. Preprocessing for feature extraction is done using discrete wavelet transform (DWT). The time domain and relative wavelet energy (RWE) features calculated from the sub-bands are given as a set of input to the neural network. Another set of feature used independently for training the network is the wavelet entropy (WE). The FFNN achieved a classification accuracy of 100% and PNN gave an accuracy of 58.75% with time domain and wavelet energy as the input features. With wavelet entropy as the input feature, FFNN further showed 98.75% classification accuracy while PNN gave an accuracy of only 46.5%. The results indicate that FFNN with the given input features is more suitable for the classification of EEG signals with mood changing depressive disorders.

A model for continuous monitoring of patients with major depression in short and long term periods

BACKGROUND AND OBJECTIVE

Major depressive disorder causes more human suffering than any other disease affecting humankind. It has a high prevalence and it is predicted that it will be among the three leading causes of disease burden by 2030. The prevalence of depression, all of its social and personal costs, and its recurrent characteristics, put heavy constraints on the ability of the public healthcare system to provide sufficient support for patients with depression. In this research, a model for continuous monitoring and tracking of depression in both short-term and long-term periods is presented. This model is based on a new qualitative reasoning approach.

METHOD

This paper describes the patient assessment unit of a major depression monitoring system that has three modules: a patient progress module, based on a qualitative reasoning model; an analysis module, based on expert knowledge and a rules-based system; and the communication module. These modules base their reasoning mainly on data of the patient's mood and life events that are obtained from the patient's responses to specific questionnaires (PHQ-9, M.I.N.I. and Brugha). The patient assessment unit provides synthetic and useful information for both patients and physicians, keeps them informed of the progress of patients, and alerts them in the case of necessity.

RESULTS

A set of hypothetical patients has been defined based on clinically possible cases in order to perform a complete scenario evaluation. The results that have been verified by psychiatrists suggest the utility of the platform.

CONCLUSION

The proposed major depression monitoring system takes advantage of current technologies and facilitates more frequent follow-up of the progress of patients during their home stay after being diagnosed with depression by a psychiatrist.