Invited Talk, NVIDIA GTC 2021, Virtual Conference
Speaker at Nvidia GTC 2021. Delivered a talk on the application of deep learning in the medical imaging and presented a novel loss function for robust histopathology image analysis.
Conference proceedings talk, IEEE International Symposium on Biomedical Imaging (ISBI), 2021, Virtual Conference
The accuracy of deep learning classifiers trained using the cross entropy loss function suffers even when a fraction of training labels are wrong or input images are uninformative. Training images and labels for computational pathology are often noisy due to the difficulty in signal localization and certain disease classifications being subjective discretization of the underlying continuums of disease conditions. For training classifiers robust to input and label noise, we propose a modified and sample-specific version of generalized cross entropy loss. We take advantage of the bootstrapping properties in deep learning models in order to design loss functions that are aware of the difficulty of classifying individual samples, due to either the label noise or the lack of a strong visual signal. We carry out extensive experiments to validate our approach by comparing against the models trained using other loss functions. The superior performance of our methodology in all the experiments support the requirement of using robust loss functions for histology image classification.
FDP, Virtual Workshop, Mumbai, India
AICTE Sponsored Faculty Development Program(FDP), on the topic ”The confluence of deep learning in histopathology images”, Virtual Program 2021
Tutorial, Tata Memorial Centre Advanced Centre for Treatment, Research and Education in Cancer , Navi Mumbai, India
As a workshop speaker, introducing the convergence of deep learning for automatic histology image anlysis. Over 100 participants from diverse fields inlcuding medical practioners, doctors, engineering graduates attended the workshop.
Conference Proceedings, Hotel Grand Hyatt, Kochi, India
Conference proceedings on our paper: “Improving Histopathology Classification using Learnable Preprocessing”
Abstract:
A deep learning classifier trained on a source dataset often performs poorly on a target dataset, even for the same classification task, due to the differences in the distributions of the two datasets. In histopathology, the problem of dataset bias is even more severe due to the differences in specific tissue preparation and imaging set ups across patient cohorts. With the objective of improving the generalization across datasets, we propose a set of learnable preprocessing operations - an approach that has not been extensively explored - for a supervised deep learning framework that can be trained separately or together with the rest of the neural network. Through preprocessing, the data from a target domain is transformed before being fed to a classification module trained on the source domain to increase the overlap of the former's distribution with that of the latter. Through an extensive set of experiments on histopathology and face datasets, we show the particular and general utility of the proposed preprocessing operations for domain adaptation and compare it to previous approaches.
Conference Proceedings, Phuket Graceland Resort, Phuket, Thailand
Conference proceedings on our paper: “Time frequency analysis: a sparse S transform approach”
Abstract:
S transform, which is a powerful time frequency analysis method, has found applications in diverse areas of science and technology. The computational load offered by the S transform increases with increase in the length of the time series which is analysed. In an endeavour to reduce the computational load for time series which is sparse in the frequency domain, a new method for S transform computation is proposed in this paper. The new method uses an efficient search method to identify significant frequency indices and computes the S transform only at the selected frequency indices, thus reducing the computational burden. A simulation study has been carried out to test the efficiency of the proposed method for analytic and real-life signals. The proposed scheme has been shown to provide good signal reconstruction accuracy at a reduced computational load.