results for au:Venkataramani_R in:cs
Apr 21 2017 cs.CV
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.
Dec 09 2016 cs.CV
Typical convolutional neural networks (CNNs) have several millions of parameters and require a large amount of annotated data to train them. In medical applications where training data is hard to come by, these sophisticated machine learning models are difficult to train. In this paper, we propose a method to reduce the inherent complexity of CNNs during training by exploiting the significant redundancy that is noticed in the learnt CNN filters. Our method relies on finding a small set of filters and mixing coefficients to derive every filter in each convolutional layer at the time of training itself, thereby reducing the number of parameters to be trained. We consider the problem of 3D lung nodule segmentation in CT images and demonstrate the effectiveness of our method in achieving good results with only few training examples.
The problem of channel shortening equalization for optimal detection in ISI channels is considered. The problem is to choose a linear equalizer and a partial response target filter such that the combination produces the best detection performance. Instead of using the traditional approach of MMSE equalization, we directly seek all equalizer and target pairs that yield optimal detection performance in terms of the sequence or symbol error rate. This leads to a new notion of a posteriori equivalence between the equalized and target channels with a simple characterization in terms of their underlying probability distributions. Using this characterization we show the surprising existence an infinite family of equalizer and target pairs for which any maximum a posteriori (MAP) based detector designed for the target channel is simultaneously MAP optimal for the equalized channel. For channels whose input symbols have equal energy, such as q-PSK, the MMSE equalizer designed with a monic target constraint yields a solution belonging to this optimal family of designs. Although, these designs produce IIR target filters, the ideas are extended to design good FIR targets. For an arbitrary choice of target and equalizer, we derive an expression for the probability of sequence detection error. This expression is used to design optimal FIR targets and IIR equalizers and to quantify the FIR approximation penalty.