Inconsistency between Human Observation and Deep Learning Models: Assessing Validity of Postmortem Computed Tomography Diagnosis of Drowning.

Drowning diagnosis is a complicated process in the autopsy, even with the assistance of autopsy imaging and the on-site information from where the body was found. Previous studies have developed well-performed deep learning (DL) models for drowning diagnosis. However, the validity of the DL models was not assessed, raising doubts about whether the learned features accurately represented the medical findings observed by human experts. In this paper, we assessed the medical validity of DL models that had achieved high classification performance for drowning diagnosis. This retrospective study included autopsy cases aged 8-91 years who underwent postmortem computed tomography between 2012 and 2021 (153 drowning and 160 non-drowning cases). We first trained three deep learning models from a previous work and generated saliency maps that highlight important features in the input. To assess the validity of models, pixel-level annotations were created by four radiological technologists and further quantitatively compared with the saliency maps. All the three models demonstrated high classification performance with areas under the receiver operating characteristic curves of 0.94, 0.97, and 0.98, respectively. On the other hand, the assessment results revealed unexpected inconsistency between annotations and models' saliency maps. In fact, each model had, respectively, around 30%, 40%, and 80% of irrelevant areas in the saliency maps, suggesting the predictions of the DL models might be unreliable. The result alerts us in the careful assessment of DL tools, even those with high classification performance.

How intra-source imbalanced datasets impact the performance of deep learning for COVID-19 diagnosis using chest X-ray images.

Over the past decade, the use of deep learning has been widely increasing in the medical image diagnosis field. Deep learning-based methods' (DLMs) performance strongly relies on training data. Therefore, researchers often focus on collecting as much data as possible from different medical facilities or developing approaches to avoid the impact of inter-category imbalance (ICI), which means a difference in data quantity among categories. However, due to the ICI within each medical facility, medical data are often isolated and acquired in different settings among medical facilities, known as the issue of intra-source imbalance (ISI) characteristic. This imbalance also impacts the performance of DLMs but receives negligible attention. In this study, we study the impact of the ISI on DLMs by comparison of the version of a deep learning model that was trained separately by an intra-source imbalanced chest X-ray (CXR) dataset and an intra-source balanced CXR dataset for COVID-19 diagnosis. The finding is that using the intra-source imbalanced dataset causes a serious training bias, although the dataset has a good inter-category balance. In contrast, the deep learning model performed a reliable diagnosis when trained on the intra-source balanced dataset. Therefore, our study reports clear evidence that the intra-source balance is vital for training data to minimize the risk of poor performance of DLMs.

Deep Learning-Based Diagnosis of Fatal Hypothermia Using Post-Mortem Computed Tomography.

In forensic medicine, fatal hypothermia diagnosis is not always easy because findings are not specific, especially if traumatized. Post-mortem computed tomography (PMCT) is a useful adjunct to the cause-of-death diagnosis and some qualitative image character analysis, such as diffuse hyperaeration with decreased vascularity or pulmonary emphysema, have also been utilized for fatal hypothermia. However, it is challenging for inexperienced forensic pathologists to recognize the subtle differences of fatal hypothermia in PMCT images. In this study, we developed a deep learning-based diagnosis system for fatal hypothermia and explored the possibility of being an alternative diagnostic for forensic pathologists. An in-house dataset of forensic autopsy proven samples was used for the development and performance evaluation of the deep learning system. We used the area under the receiver operating characteristic curve (AUC) of the system for evaluation, and a human-expert comparable AUC value of 0.905, sensitivity of 0.948, and specificity of 0.741 were achieved. The experimental results clearly demonstrated the usefulness and feasibility of the deep learning system for fatal hypothermia diagnosis.

Comments on "Convergence Analysis of Adaptive Exponential Functional Link Network".

This article is to comment on the derivation of the weight-update stability of in-parameter-linear nonlinear learning system with the gradient descent learning rule in the above article. Our comments are not to disqualify the commented article's whole contribution; however, the issues should be pointed out to avoid their proliferation.

Letter on Convergence of In-Parameter-Linear Nonlinear Neural Architectures With Gradient Learnings.

This letter summarizes and proves the concept of bounded-input bounded-state (BIBS) stability for weight convergence of a broad family of in-parameter-linear nonlinear neural architectures (IPLNAs) as it generally applies to a broad family of incremental gradient learning algorithms. A practical BIBS convergence condition results from the derived proofs for every individual learning point or batches for real-time applications.

Novelty detection-based approach for Alzheimer's disease and mild cognitive impairment diagnosis from EEG.

Alzheimer's disease is diagnosed via means of daily activity assessment. The EEG recording evaluation is a supporting tool that can assist the practitioner to recognize the illness, especially in the early stages. This paper presents a new approach for detecting Alzheimer's disease and potentially mild cognitive impairment according to the measured EEG records. The proposed method evaluates the amount of novelty in the EEG signal as a feature for EEG record classification. The novelty is measured from the parameters of EEG signal adaptive filtration. A linear neuron with gradient descent adaptation was used as the filter in predictive settings. The extracted feature (novelty measure) is later classified to obtain Alzheimer's disease diagnosis. The proposed approach was cross-validated on a dataset containing EEG records of 59 patients suffering from Alzheimer's disease; seven patients with mild cognitive impairment (MCI) and 102 controls. The results of cross-validation yield 90.73% specificity and 89.51% sensitivity. The proposed method of feature extraction from EEG is completely new and can be used with any classifier for the diagnosis of Alzheimer's disease from EEG records.

A Deep Learning Aided Drowning Diagnosis for Forensic Investigations using Post-Mortem Lung CT Images.

Feasibility of computer-aided diagnosis (CAD) systems has been demonstrated in the field of medical image diagnosis. Especially, deep learning based CAD systems showed high performance thanks to its capability of image recognition. However, there is no CAD system developed for post-mortem imaging diagnosis and thus it is still unclear if the CAD system is effective for this purpose. Particulally, the drowning diagnosis is one of the most difficult tasks in the field of forensic medicine because findings of the post-mortem image diagnosis are not specific. To address this issue, we develop a CAD system consisting of a deep convolution neural network (DCNN) to classify post-mortem lung computed tomography (CT) images into two categories of drowning and non-drowning cases. The DCNN was trained by means of transfer learning and performance evaluation was conducted by 10-fold cross validation using 140 drowning cases and 140 non-drowning cases of the CT images. The area under the receiver operating characteristic curve (AUC-ROC) for the DCNN was achieved 0.88 in average. This high performance clearly demonstrated that the proposed DCNN based CAD system has a potential for post-mortem image diagnosis of drowning.

Learning Entropy as a Learning-Based Information Concept.

Recently, a novel concept of a non-probabilistic novelty detection measure, based on a multi-scale quantification of unusually large learning efforts of machine learning systems, was introduced as learning entropy (LE). The key finding with LE is that the learning effort of learning systems is quantifiable as a novelty measure for each individually observed data point of otherwise complex dynamic systems, while the model accuracy is not a necessary requirement for novelty detection. This brief paper extends the explanation of LE from the point of an informatics approach towards a cognitive (learning-based) information measure emphasizing the distinction from Shannon's concept of probabilistic information. Fundamental derivations of learning entropy and of its practical estimations are recalled and further extended. The potentials, limitations, and, thus, the current challenges of LE are discussed.

Dosimetric evaluation of MLC-based dynamic tumor tracking radiotherapy using digital phantom: Desired setup margin for tracking radiotherapy.

The purpose of this study is to evaluate the dosimetric impact of the margin on the multileaf collimator-based dynamic tumor tracking plan. Furthermore, an equivalent setup margin (EM) of the tracking plan was determined according to the gated plan. A 4-dimensional extended cardiac-torso was used to create 9 digital phantom datasets of different tumor diameters (TDs) of 1, 3, and 5 cm and motion ranges (MRs) of 1, 2, and 3 cm. For each dataset, respiratory gating (30% to 70% phase) and tumor tracking treatment plans were prepared using 8-field 3-dimensional conformal radiation therapy by 4-dimensional dose calculation. The total lung V20 was calculated to evaluate the dosimetric impact for each case and to estimate the EM with the same impact on lung V20 obtained with the gating plan with a setup margin of 5 mm. The EMs for {TD = 1 cm, MR = 1 cm}, {TD = 1 cm, MR = 2 cm}, and {TD = 1 cm, MR = 3 cm} were estimated as 5.00, 4.16, and 4.24 mm, respectively. The EMs for {TD = 5 cm, MR = 1 cm}, {TD = 5 cm, MR = 2 cm}, and {TD = 5 cm, MR = 3 cm} were estimated as 4.24 mm, 6.35 mm, and 7.49 mm, respectively. This result showed that with a larger MR, the EM was found to be increased. In addition, with a larger TD, the EM became smaller. Our result showing the EMs provided the desired accuracy for multileaf collimator-based dynamic tumor tracking radiotherapy.

An Approach to Stable Gradient-Descent Adaptation of Higher Order Neural Units.

Stability evaluation of a weight-update system of higher order neural units (HONUs) with polynomial aggregation of neural inputs (also known as classes of polynomial neural networks) for adaptation of both feedforward and recurrent HONUs by a gradient descent method is introduced. An essential core of the approach is based on the spectral radius of a weight-update system, and it allows stability monitoring and its maintenance at every adaptation step individually. Assuring the stability of the weight-update system (at every single adaptation step) naturally results in the adaptation stability of the whole neural architecture that adapts to the target data. As an aside, the used approach highlights the fact that the weight optimization of HONU is a linear problem, so the proposed approach can be generally extended to any neural architecture that is linear in its adaptable parameters.

A fast neural network approach to predict lung tumor motion during respiration for radiation therapy applications.

During radiotherapy treatment for thoracic and abdomen cancers, for example, lung cancers, respiratory motion moves the target tumor and thus badly affects the accuracy of radiation dose delivery into the target. A real-time image-guided technique can be used to monitor such lung tumor motion for accurate dose delivery, but the system latency up to several hundred milliseconds for repositioning the radiation beam also affects the accuracy. In order to compensate the latency, neural network prediction technique with real-time retraining can be used. We have investigated real-time prediction of 3D time series of lung tumor motion on a classical linear model, perceptron model, and on a class of higher-order neural network model that has more attractive attributes regarding its optimization convergence and computational efficiency. The implemented static feed-forward neural architectures are compared when using gradient descent adaptation and primarily the Levenberg-Marquardt batch algorithm as the ones of the most common and most comprehensible learning algorithms. The proposed technique resulted in fast real-time retraining, so the total computational time on a PC platform was equal to or even less than the real treatment time. For one-second prediction horizon, the proposed techniques achieved accuracy less than one millimeter of 3D mean absolute error in one hundred seconds of total treatment time.