Hariharan Ravishankar

Understanding the Mechanisms of Deep Transfer Learning for Medical Images

The ability to automatically learn task specific feature representations has led to a huge success of deep learning methods. When large training data is scarce, such as in medical imaging problems, transfer learning has been very effective. In this paper, we systematically investigate the process of transferring a Convolutional Neural Network, trained on ImageNet images to perform image classification, to kidney detection problem in ultrasound images. We study how the detection performance depends on the extent of transfer. We show that a transferred and tuned CNN can outperform a state-of-the-art feature engineered pipeline and a hybridization of these two techniques achieves 20 % higher performance. We also investigate how the evolution of intermediate response images from our network. Finally, we compare these responses to state-of-the-art image processing filters in order to gain greater insight into how transfer learning is able to effectively manage widely varying imaging regimes.

Learning and Incorporating Shape Models for Semantic Segmentation

Semantic segmentation has been popularly addressed using Fully convolutional networks (FCN) (e.g. U-Net) with impressive results and has been the forerunner in recent segmentation challenges. However, FCN approaches do not necessarily incorporate local geometry such as smoothness and shape, whereas traditional image analysis techniques have benefitted greatly by them in solving segmentation and tracking problems. In this work, we address the problem of incorporating shape priors within the FCN segmentation framework. We demonstrate the utility of such a shape prior in robust handling of scenarios such as loss of contrast and artifacts. Our experiments show \(\approx 5\%\) improvement over U-Net for the challenging problem of ultrasound kidney segmentation.

Hybrid approach for automatic segmentation of fetal abdomen from ultrasound images using deep learning

In this paper, we propose a hybrid approach combining traditional texture analysis methods with deep learning for the automatic detection and measurement of abdominal contour from 2-D fetal ultrasound images. Following a learning-based procedure for region of interest (ROI) localization to segment the abdominal boundary, we show that convolutional neural networks (CNNs) outperform other state-of-the-art texture features and conventional classifiers, in addressing the binary classification problem of distinguishing between abdomen versus non-abdomen regions. However, we obtain significantly better segmentation results in identifying the best ROI containing fetal abdomen, when the predictions from CNN are combined with those from gradient boosting machine (GBM) using histogram of oriented gradient (HOG) features. We trained our method on a set of 70 images and tested them on another distinct set of 70 images. We obtained a mean DICE similarity coefficient of 0.90, which shows excellent overlap with the ground truth. We report that the mean computed gestational age difference between our segmentation results and the ground truth, is within two weeks for 90% (and within one week for 70%) of the testing cases.

An early respiratory distress detection method with Markov models.

A method for early detection of respiratory distress in hospitalized patients which is based on a multi-parametric analysis of respiration rate (RR) and pulse oximetry (SpO2) data trends to ascertain patterns of patient instability pertaining to respiratory distress is described. Current practices of triggering caregiver alerts are based on simple numeric threshold breaches of SpO2. The pathophysiological patterns of respiratory distress leading to in-hospital deaths are much more complex to be detected by numeric thresholds. Our pattern detection algorithm is based on a Markov model framework based on multi-parameter pathophysiological patterns of respiratory distress, and triggers in a timely manner and prior to the violation of SpO2 85-90% threshold, providing additional lead time to attempt to reverse the deteriorating state of the patient. We present the performance of the algorithm on MIMIC II dataset resulting in true positive rate of 92% and false positive rate of 6%.

Automated kidney morphology measurements from ultrasound images using texture and edge analysis

In a typical ultrasound scan, a sonographer measures Kidney morphology to assess renal abnormalities. Kidney morphology can also help to discriminate between chronic and acute kidney failure. The caliper placements and volume measurements are often time consuming and an automated solution will help to improve accuracy, repeatability and throughput. In this work, we developed an automated Kidney morphology measurement solution from long axis Ultrasound scans. Automated kidney segmentation is challenging due to wide variability in kidney shape, size, weak contrast of the kidney boundaries and presence of strong edges like diaphragm, fat layers. To address the challenges and be able to accurately localize and detect kidney regions, we present a two-step algorithm that makes use of edge and texture information in combination with anatomical cues. First, we use an edge analysis technique to localize kidney region by matching the edge map with predefined templates. To accurately estimate the kidney morphology, we use textural information in a machine learning algorithm framework using Haar features and Gradient boosting classifier. We have tested the algorithm on 45 unseen cases and the performance against ground truth is measured by computing Dice overlap, % error in major and minor axis of kidney. The algorithm shows successful performance on 80% cases.

Joint Deep Learning of Foreground, Background and Shape for Robust Contextual Segmentation

Encouraged by the success of CNNs in classification problems, CNNs are being actively applied to image-wide prediction problems such as segmentation, optic flow, reconstruction, restoration etc. These approaches fall under the category of fully convolutional networks [FCN] and have been very successful in bringing contexts into learning for image analysis. In this work, we address the problem of segmentation from medical images. Segmentation or object delineation from medical images/volumes is a fundamental step for subsequent quantification tasks key to diagnosis. Semantic segmentation has been popularly addressed using FCN (e.g. U-NET) with impressive results and has been the fore runner in recent segmentation challenges. However, there are a few drawbacks of FCN approaches which recent works have tried to address. Firstly, local geometry such as smoothness and shape are not reliably captured. Secondly, spatial context captured by FCNs while giving the advantage of a richer representation carries the intrinsic drawback of overfitting, and is quite sensitive to appearance and shape changes. To handle above issues, in this work, we propose a hybrid of generative modeling of image formation to jointly learn the triad of foreground (F), background (B) and shape (S). Such generative modeling of F, B, S would carry the advantages of FCN in capturing contexts. Further we expect the approach to be useful under limited training data, results easy to interpret, and enable easy transfer of learning across segmentation problems. We present \({\sim }8\%\) improvement over state of art FCN approaches for US kidney segmentation and while achieving comparable results on CT lung nodule segmentation.

Recursive feature elimination for biomarker discovery in resting-state functional connectivity

Biomarker discovery involves finding correlations between features and clinical symptoms to aid clinical decision. This task is especially difficult in resting state functional magnetic resonance imaging (rs-fMRI) data due to low SNR, high-dimensionality of images, inter-subject and intra-subject variability and small numbers of subjects compared to the number of derived features. Traditional univariate analysis suffers from the problem of multiple comparisons. Here, we adopt an alternative data-driven method for identifying population differences in functional connectivity. We propose a machine-learning approach to down-select functional connectivity features associated with symptom severity in mild traumatic brain injury (mTBI). Using this approach, we identified functional regions with altered connectivity in mTBI. including the executive control, visual and precuneus networks. We compared functional connections at multiple resolutions to determine which scale would be more sensitive to changes related to patient recovery. These modular network-level features can be used as diagnostic tools for predicting disease severity and recovery profiles.