Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

An enhanced approach to the robust discriminant analysis and class sparsity based embedding.

In Neural networks : the official journal of the International Neural Network Society

In recent times, feature extraction attracted much attention in machine learning and pattern recognition fields. This paper extends and improves a scheme for linear feature extraction that can be used in supervised multi-class classification problems. Inspired by recent frameworks for robust sparse LDA and Inter-class sparsity, we propose a unifying criterion able to retain the advantages of these two powerful linear discriminant methods. We introduce an iterative alternating minimization scheme in order to estimate the linear transformation and the orthogonal matrix. The linear transformation is efficiently updated via the steepest descent gradient technique. The proposed framework is generic in the sense that it allows the combination and tuning of other linear discriminant embedding methods. We used our proposed method to fine tune the linear solutions delivered by two recent linear methods: RSLDA and RDA_FSIS. Experiments have been conducted on public image datasets of different types including objects, faces, and digits. The proposed framework compared favorably with several competing methods.

Khoder A, Dornaika F


Discriminant model, Feature extraction, Image categorization, Inter-class sparsity, Linear embedding, Supervised learning

Radiology Radiology

Transfer learning for predicting conversion from mild cognitive impairment to dementia of Alzheimer's type based on a three-dimensional convolutional neural network.

In Neurobiology of aging ; h5-index 69.0

Dementia of Alzheimer's type (DAT) is associated with devastating and irreversible cognitive decline. Predicting which patients with mild cognitive impairment (MCI) will progress to DAT is an ongoing challenge in the field. We developed a deep learning model to predict conversion from MCI to DAT. Structural magnetic resonance imaging scans were used as input to a 3-dimensional convolutional neural network. The 3-dimensional convolutional neural network was trained using transfer learning; in the source task, normal control and DAT scans were used to pretrain the model. This pretrained model was then retrained on the target task of classifying which MCI patients converted to DAT. Our model resulted in 82.4% classification accuracy at the target task, outperforming current models in the field. Next, we visualized brain regions that significantly contribute to the prediction of MCI conversion using an occlusion map approach. Contributory regions included the pons, amygdala, and hippocampus. Finally, we showed that the model's prediction value is significantly correlated with rates of change in clinical assessment scores, indicating that the model is able to predict an individual patient's future cognitive decline. This information, in conjunction with the identified anatomical features, will aid in building a personalized therapeutic strategy for individuals with MCI.

Bae Jinhyeong, Stocks Jane, Heywood Ashley, Jung Youngmoon, Jenkins Lisanne, Hill Virginia, Katsaggelos Aggelos, Popuri Karteek, Rosen Howie, Beg M Faisal, Wang Lei


Convolutional neural network, “Dementia of Alzheimers type”, Magnetic resonance imaging, Mild cognitive impairment, Predictive modeling

Radiology Radiology

A neural network to predict the knee adduction moment in patients with osteoarthritis using anatomical landmarks obtainable from 2D video analysis.

In Osteoarthritis and cartilage ; h5-index 62.0

OBJECTIVE : The knee adduction moment (KAM) can inform treatment of medial knee osteoarthritis; however, measuring the KAM requires an expensive gait analysis laboratory. We evaluated the feasibility of predicting the peak KAM during natural and modified walking patterns using the positions of anatomical landmarks that could be identified from video analysis.

METHOD : Using inverse dynamics, we calculated the KAM for 86 individuals (64 with knee osteoarthritis, 22 without) walking naturally and with foot progression angle modifications. We trained a neural network to predict the peak KAM using the 3-dimensional positions of 13 anatomical landmarks measured with motion capture (3D neural network). We also trained models to predict the peak KAM using 2-dimensional subsets of the dataset to simulate 2-dimensional video analysis (frontal and sagittal plane neural networks). Model performance was evaluated on a held-out, 8-person test set that included steps from all trials.

RESULTS : The 3D neural network predicted the peak KAM for all test steps with r2=0.78. This model predicted individuals' average peak KAM during natural walking with r2=0.86 and classified which 15° foot progression angle modifications reduced the peak KAM with accuracy=0.85. The frontal plane neural network predicted peak KAM with similar accuracy (r2=0.85) to the 3D neural network, but the sagittal plane neural network did not (r2=0.14).

CONCLUSION : Using the positions of anatomical landmarks from motion capture, a neural network accurately predicted the peak KAM during natural and modified walking. This study demonstrates the feasibility of measuring the peak KAM using positions obtainable from 2D video analysis.

Boswell Melissa A, Uhlrich Scott D, Kidziński Łukasz, Thomas Kevin, Kolesar Julie A, Gold Garry E, Beaupre Gary S, Delp Scott L


Gait, Knee Adduction Moment, Machine Learning, Neural Network, Osteoarthritis

General General

Deep learning enables the automation of grading histological tissue engineered cartilage images for quality control standardization.

In Osteoarthritis and cartilage ; h5-index 62.0

OBJECTIVE : To automate the grading of histological images of engineered cartilage tissues using deep learning.

METHODS : Cartilaginous tissues were engineered from various cell sources. Safranin O and fast green stained histological images of the tissues were graded for chondrogenic quality according to the Modified Bern Score, which ranks images on a scale from zero to six according to the intensity of staining and cell morphology. The whole images were tiled, and the tiles were graded by two experts and grouped into four categories with the following grades: 0, 1-2, 3-4, and 5-6. Deep learning was used to train models to classify images into these histological score groups. Finally, the tile grades per donor were averaged. The root mean square errors (RMSEs) were calculated between each user and the model.

RESULTS : Transfer learning using a pretrained DenseNet model was selected. The RMSEs of the model predictions and 95% confidence intervals were 0.49 (0.37, 0.61) and 0.78 (0.57, 0.99) for each user, which was in the same range as the inter-user RMSE of 0.71 (0.51, 0.93).

CONCLUSION : Using supervised deep learning, we could automate the scoring of histological images of engineered cartilage and achieve results with errors comparable to inter-user error. Thus, the model could enable the automation and standardization of assessments currently used for experimental studies as well as release criteria that ensure the quality of manufactured clinical grafts and compliance with regulatory requirements.

Power Laura, Acevedo Lina, Yamashita Rikiya, Rubin Daniel, Martin Ivan, Barbero Andrea


Machine learning, convolutional neural networks, histological score, quality controls, regenerative medicine, transfer learning

Cardiology Cardiology

Importance of systematic right ventricular assessment in cardiac resynchronization therapy candidates: a machine-learning approach.

In Journal of the American Society of Echocardiography : official publication of the American Society of Echocardiography

BACKGROUND : Despite all having systolic heart failure and broad QRS, patients screened for cardiac resynchronization therapy (CRT) are highly heterogeneous, and it remains extremely challenging to predict the impact of the device on left ventricular (LV) function and outcomes.

OBJECTIVES : We sought to evaluate the relative impact of clinical, electrocardiographic, and echocardiographic data on the left ventricular (LV) remodeling and prognosis of CRT-candidates by the application of machine learning (ML) approaches.

METHODS : 193 patients with systolic heart failure undergoing CRT according to current recommendations were prospectively included in this multicentre study. We used a combination of the Boruta algorithm and random forest methods to identify features predicting both CRT volumetric response and prognosis. The model performance was tested by the area under the receiver operating curve (AUC). We also applied the K-medoid method to identify clusters of phenotypically similar patients.

RESULTS : From 28 clinical, electrocardiographic, and echocardiographic-derived variables, 16 features were predictive of CRT response, and 11 features were predictive of prognosis. Among the predictors of CRT-response, 8 variables (50%) pertained to right ventricular (RV) size or function. Tricuspid annular plane systolic excursion was the main feature associated with prognosis. The selected features were associated with a particularly good prediction of both CRT response (AUC 0.81, 95% CI: 0.74-0.87) and outcomes (AUC 0.84, 95% CI: 0.75-0.93). An unsupervised ML approach allowed the identifications of two phenogroups of patients who differed significantly in clinical variables and parameters of biventricular size, and RV function. The two phenogroups had significant different prognosis (HR 4.70, 95% CI: 2.1-10.0, p<0.0001; log-rank p<0.0001).

CONCLUSIONS : Machine learning can reliably identify clinical and echocardiographic features associated with CRT-response and prognosis. The evaluation of both RV-size and function parameters has pivotal importance for the risk stratification of CRT-candidates and should be systematically assessed in patients undergoing CRT.

Galli E, Le Rolle V, Smiseth O A, Duchenne J, Aalen J M, Larsen C K, Sade E, Hubert A, Anilkumar S, Penicka M, Cecilia Linde, Leclercq C, Hernandez A, Voigt J-U, Donal E


cardiac resynchronization therapy, heart failure, machine learning, right ventricle

General General

Interpreting a Recurrent Neural Network's Predictions of ICU Mortality Risk.

In Journal of biomedical informatics ; h5-index 55.0

Deep learning has demonstrated success in many applications; however, their use in healthcare has been limited due to the lack of transparency into how they generate predictions. Algorithms such as Recurrent Neural Networks (RNNs) when applied to Electronic Medical Records (EMR) introduce additional barriers to transparency because of the sequential processing of the RNN and the multi-modal nature of EMR data. This work seeks to improve transparency by: 1) introducing Learned Binary Masks (LBM) as a method for identifying which EMR variables contributed to an RNN model's risk of mortality (ROM) predictions for critically ill children; and 2) applying KernelSHAP for the same purpose. Given an individual patient, LBM and KernelSHAP both generate an attribution matrix that shows the contribution of each input feature to the RNN's sequence of predictions for that patient. Attribution matrices can be aggregated in many ways to facilitate different levels of analysis of the RNN model and its predictions. Presented are three methods of aggregations and analyses: 1) over volatile time periods within individual patient predictions, 2) over populations of ICU patients sharing specific diagnoses, and 3) across the general population of critically ill children.

Ho Long V, Aczon Melissa, Ledbetter David, Wetzel Randall


Deep Learning, Electronic Medical Records, Feature Attribution, Feature Importance, Model Interpretation, Recurrent Neural Networks