Subhashis Banerjee
@uu.se
Assistant Vice President – Data Scientist
EDUCATION
Ph.D. (Tech.), in Computer Science and Engineering
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Subhashis Banerjee, Fredrik Nysjö, Dimitrios Toumpanakis, Ashis Kumar Dhara, Johan Wikström, and Robin Strand
Springer Science and Business Media LLC
AbstractRadiological imaging to examine intracranial blood vessels is critical for preoperative planning and postoperative follow-up. Automated segmentation of cerebrovascular anatomy from Time-Of-Flight Magnetic Resonance Angiography (TOF-MRA) can provide radiologists with a more detailed and precise view of these vessels. This paper introduces a domain generalized artificial intelligence (AI) solution for volumetric monitoring of cerebrovascular structures from multi-center MRAs. Our approach utilizes a multi-task deep convolutional neural network (CNN) with a topology-aware loss function to learn voxel-wise segmentation of the cerebrovascular tree. We use Decorrelation Loss to achieve domain regularization for the encoder network and auxiliary tasks to provide additional regularization and enable the encoder to learn higher-level intermediate representations for improved performance. We compare our method to six state-of-the-art 3D vessel segmentation methods using retrospective TOF-MRA datasets from multiple private and public data sources scanned at six hospitals, with and without vascular pathologies. The proposed model achieved the best scores in all the qualitative performance measures. Furthermore, we have developed an AI-assisted Graphical User Interface (GUI) based on our research to assist radiologists in their daily work and establish a more efficient work process that saves time.
Swagata Kundu, Subhashis Banerjee, Eva Breznik, Dimitrios Toumpanakis, Johan Wikstrom, Robin Strand, and Ashis Kumar Dhara
Springer Science and Business Media LLC
Swagata Kundu, Subhashis Banerjee, Dimitrios Toumpanakis, Johan Wikstrom, Robin Strand, and Ashis Kumar Dhara
Springer Nature Switzerland
Subhashis Banerjee and Robin Strand
Springer Nature Switzerland
Subhashis Banerjee and Robin Strand
SPIE
This paper presents a novel solution for catastrophic forgetting in lifelong learning (LL) using Dynamic Convolution Neural Network (Dy-CNN). The proposed dynamic convolution layer can adapt convolution filters by learning kernel coefficients or weights based on the input image. The suitability of the proposed Dy-CNN in a lifelong sequential learning-based scenario with multi-modal MR images is experimentally demonstrated for the segmentation of Glioma tumors from multi-modal MR images. Experimental results demonstrated the superiority of the Dy-CNN-based segmenting network in terms of learning through multi-modal MRI images and better convergence of lifelong learning-based training.
Subhashis Banerjee, Dimitrios Toumpanakis, Ashis K. Dhara, Johan Wikström, and Robin Strand
SPIE
This paper presents a weakly supervised deep convolutional neural network-based approach to perform voxel-level 3D registration between subsequent follow-up MRI scans of the same patient. To handle the large deformation in the surrounding brain tissues due to the tumor’s mass effect we proposed curriculum learning-based training for the network. Weak supervision helps the network to concentrate more focus on the tumor region and resection cavity through a saliency detection network. Qualitative and quantitative experimental results show the proposed registration network outperformed two popular state-of-the-art methods.
Subhash Chandra Pal, Subhashis Banerjee, Dimitrios Toumpanakis, Johan Wikström, Robin Strand, and Ashis Kumar Dhara
IEEE
Arterial cerebral vessel assessment is critical for the diagnosis of patients with cerebrovascular disease e.g., hypertension, Intracranial aneurysms, and dementia. Magnetic resonance angiography is a primary imaging technique for diagnosing cerebrovascular diseases. There are many Convolutional neural networks (CNN) based methods for cerebral vessel segmentation but lack to identify the target vessels and understand the arterial tree structure for diagnosis and endovascular surgical planning. In the present study, we generated annotations for major vessel segmentation and analyzed fully automatic segmentation of major vessels using state-of-the-art U-Net based deep learning models. Computer-aided major cerebral vessel segmentation incorporated into clinical practice may help speed up the diagnosis of time-critical vessel anomalies and help find important bio-markers for neurological dysfunction. We validated and compared U-Net based models for volumetric segmentation and prediction of cerebral arteries and it could be done in real-time without any image pre-processing.
Subhashis Banerjee and Sushmita Mitra
Elsevier
Subhashis Banerjee, Dimitrios Toumpanakis, Ashis Kumar Dhara, Johan Wikstrom, and Robin Strand
IEEE
This paper presents a topology-aware learning strategy for volumetric segmentation of intracranial cerebrovascular structures. We propose a multi-task deep CNN along with a topology-aware loss function for this purpose. Along with the main task (i.e. segmentation), we train the model to learn two related auxiliary tasks viz. learning the distance transform for the voxels on the surface of the vascular tree and learning the vessel centerline. This provides additional regularization and allows the encoder to learn higher-level intermediate representations to boost the performance of the main task. We compare the proposed method with six state-of-the-art deep learning-based 3D vessel segmentation methods, by using a public Time-Of-Flight Magnetic Resonance Angiography (TOF-MRA) dataset. Experimental results demonstrate that the proposed method has the best performance in this particular context.
Ritaban Kirtania, Subhashis Banerjee, Sayantan Laha, B. Uma Shankar, Raghunath Chatterjee, and Sushmita Mitra
MDPI AG
Glioblastoma Multiforme (GBM) is an aggressive form of glioma, exhibiting very poor survival. Genomic input, in the form of RNA sequencing data (RNA-seq), is expected to provide vital information about the characteristics of the genes that affect the Overall Survival (OS) of patients. This could have a significant impact on treatment planning. We present a new Autoencoder (AE)-based strategy for the prediction of survival (low or high) of GBM patients, using the RNA-seq data of 129 GBM samples from The Cancer Genome Atlas (TCGA). This is a novel interdisciplinary approach to integrating genomics with deep learning towards survival prediction. First, the Differentially Expressed Genes (DEGs) were selected using EdgeR. These were further reduced using correlation-based analysis. This was followed by the application of ranking with different feature subset selection and feature extraction algorithms, including the AE. In each case, fifty features were selected/extracted, for subsequent prediction with different classifiers. An exhaustive study for survival group prediction, using eight different classifiers with the accuracy and Area Under the Curve (AUC), established the superiority of the AE-based feature extraction method, called DeepSGP. It produced a very high accuracy (0.83) and AUC (0.90). Of the eight classifiers, using the extracted features by DeepSGP, the MLP was the best at Overall Survival (OS) prediction with an accuracy of 0.89 and an AUC of 0.97. The biological significance of the genes extracted by the AE were also analyzed to establish their importance. Finally, the statistical significance of the predicted output of the DeepSGP algorithm was established using the concordance index.
Subhashis Banerjee and Sushmita Mitra
IEEE
Among the different Deep Learning (DL) models, the deep Convolutional Neural Networks (CNNs) have demonstrated impressive performance in a variety of image recognition or classification tasks. Although CNNs do not require feature engineering or manual extraction of features at the input level, yet designing a suitable CNN architecture necessitates considerable expert knowledge involving enormous amount of trial-and-error activities. In this paper we attempt to automatically design a competitive CNN architecture for a given problem while consuming reasonable machine resource(s) based on a modified version of Cartesian Genetic Programming (CGP). As CGP uses only the mutation operator to generate offsprings it typically evolves slowly. We develop a new algorithm which introduces crossover to the standard CGP to generate an optimal CNN architecture. The genotype encoding scheme is changed from integer to floating-point representation for this purpose. The function genes in the nodes of the CGP are chosen as the highly functional modules of CNN. Typically CNNs use convolution and pooling, followed by activation. Rather than using each of them separately as a function gene for a node, we combine them in a novel way to construct highly functional modules. Five types of functions, called ConvBlock, average pooling, max pooling, summation, and concatenation, were considered. We test our method on an image classification dataset CIFAR10, since it is being used as the benchmark for many similar problems. Experiments demonstrate that the proposed scheme converges fast and automatically finds the competitive CNN architecture as compared to state-of-the-art solutions which require thousands of generations or GPUs involving huge computational burden.
Saumya Bhadani, Sushmita Mitra, and Subhashis Banerjee
Springer Science and Business Media LLC
Subhashis Banerjee, Sushmita Mitra, Francesco Masulli, and Stefano Rovetta
Springer Science and Business Media LLC
Abhilash Nandy, Sushovan Haldar, Subhashis Banerjee, and Sushmita Mitra
IEEE
Computer Vision nowadays uses many Deep Learning techniques in order to make the computer learn data representations from images and image sequences (as in videos). One of the important tasks in this respect is learning the similarity between two given images, which can be readily accomplished by learning a similarity criterion between the images. This can be readily accomplished using Siamese Convolutional Neural Networks (Siamese CNNs). Siamese CNNs can learn a similarity criterion between various kinds of image pairs. The paper presents a survey, which deals with the study of some remarkable papers which have used Siamese CNNs and triplet nets (which are a variation of the Siamese CNNs) in order to learn how similar are two images to one another.
Subhashis Banerjee and Sushmita Mitra
Frontiers Media SA
A novel deep learning based model called Multi-Planar Spatial Convolutional Neural Network (MPS-CNN) is proposed for effective, automated segmentation of different sub-regions viz. peritumoral edema (ED), necrotic core (NCR), enhancing and non-enhancing tumor core (ET/NET), from multi-modal MR images of the brain. An encoder-decoder type CNN model is designed for pixel-wise segmentation of the tumor along three anatomical planes (axial, sagittal, and coronal) at the slice level. These are then combined, by incorporating a consensus fusion strategy with a fully connected Conditional Random Field (CRF) based post-refinement, to produce the final volumetric segmentation of the tumor and its constituent sub-regions. Concepts, such as spatial-pooling and unpooling are used to preserve the spatial locations of the edge pixels, for reducing segmentation error around the boundaries. A new aggregated loss function is also developed for effectively handling data imbalance. The MPS-CNN is trained and validated on the recent Multimodal Brain Tumor Segmentation Challenge (BraTS) 2018 dataset. The Dice scores obtained for the validation set for whole tumor (WT :NCR/NE +ET +ED), tumor core (TC:NCR/NET +ET), and enhancing tumor (ET) are 0.90216, 0.87247, and 0.82445. The proposed MPS-CNN is found to perform the best (based on leaderboard scores) for ET and TC segmentation tasks, in terms of both the quantitative measures (viz. Dice and Hausdorff). In case of the WT segmentation it also achieved the second highest accuracy, with a score which was only 1% less than that of the best performing method.
Subhashis Banerjee, Ashis Kumar Dhara, Johan Wikstrom, and Robin Strand
IEEE
Due to the advancement of non-invasive medical imaging modalities like Magnetic Resonance Angiography (MRA), an increasing number of Intracranial Aneurysm (IA) cases are being reported in recent years. The IAs are typically treated by so-called endovascular coiling, where blood flow in the IA is prevented by embolization with a platinum coil. Accurate quantification of the IA Remnant (IAR), i.e. the volume with blood flow present post treatment is the utmost important factor in choosing the right treatment planning. This is typically done by manually segmenting the aneurysm remnant from the MRA volume. Since manual segmentation of volumetric images is a labour-intensive and error-prone process, development of an automatic volumetric segmentation method is required. Segmentation of small structures such as IA, that may largely vary in size, shape, and location is considered extremely difficult. Similar intensity distribution of IAs and surrounding blood vessels makes it more challenging and susceptible to false positive. In this paper we propose a novel 3D CNN architecture called Dual-Attention Atrous Net (DAtt-ANet), which can efficiently segment IAR volumes from MRA images by reconciling features at different scales using the proposed Parallel Atrous Unit (PAU) along with the use of self-attention mechanism for extracting fine-grained features and intra-class correlation. The proposed DAtt- ANet model is trained and evaluated on a clinical MRA image dataset of IAR consisting of 46 subjects. We compared the proposed DAtt-ANet with five state-of-the-art CNN models based on their segmentation performance. The proposed DAtt-ANet outperformed all other methods and was able to achieve a five-fold cross-validation DICE score of 0.73 ± 0.06.
Mousumi Sardar, Subhashis Banerjee, and Sushmita Mitra
Institute of Electrical and Electronics Engineers (IEEE)
Automated iris segmentation is an important component of biometric identification. The role of artificial intelligence, particularly machine learning and deep learning, has been considerable in such automated delineation strategies. Although the use of deep learning is a promising approach in recent times, some of its challenges include its high computational requirement as well as availability of large annotated training data. In this scenario, interactive learning offers a cost-effective yet efficient alternative. We introduce an interactive variant of UNet for iris segmentation, including Squeeze Expand modules, to lower training time while improving storage efficiency through a reduction in the number of parameters involved. The interactive component helps in generating the ground truth for datasets having insufficient annotated samples. The effectiveness of the model ISqEUNet is illustrated through the use of three publicly available iris databases, along with comparisons involving existing state-of-the-art methodologies.
Subhashis Banerjee, Harkirat Singh Arora, and Sushmita Mitra
Springer International Publishing
Subhashis Banerjee, Sushmita Mitra, and B. Uma Shankar
Springer International Publishing
Subhashis Banerjee, Sushmita Mitra, Francesco Masulli, and Stefano Rovetta
Springer International Publishing
Subhashis Banerjee, Sushmita Mitra, and B. Uma Shankar
Elsevier BV
Somosmita Mitra, Subhashis Banerjee, and Yoichi Hayashi
Public Library of Science (PLoS)
Medical image processing has become a major player in the world of automatic tumour region detection and is tantamount to the incipient stages of computer aided design. Saliency detection is a crucial application of medical image processing, and serves in its potential aid to medical practitioners by making the affected area stand out in the foreground from the rest of the background image. The algorithm developed here is a new approach to the detection of saliency in a three dimensional multi channel MR image sequence for the glioblastoma multiforme (a form of malignant brain tumour). First we enhance the three channels, FLAIR (Fluid Attenuated Inversion Recovery), T2 and T1C (contrast enhanced with gadolinium) to generate a pseudo coloured RGB image. This is then converted to the CIE L*a*b* color space. Processing on cubes of sizes k = 4, 8, 16, the L*a*b* 3D image is then compressed into volumetric units; each representing the neighbourhood information of the surrounding 64 voxels for k = 4, 512 voxels for k = 8 and 4096 voxels for k = 16, respectively. The spatial distance of these voxels are then compared along the three major axes to generate the novel 3D saliency map of a 3D image, which unambiguously highlights the tumour region. The algorithm operates along the three major axes to maximise the computation efficiency while minimising loss of valuable 3D information. Thus the 3D multichannel MR image saliency detection algorithm is useful in generating a uniform and logistically correct 3D saliency map with pragmatic applicability in Computer Aided Detection (CADe). Assignment of uniform importance to all three axes proves to be an important factor in volumetric processing, which helps in noise reduction and reduces the possibility of compromising essential information. The effectiveness of the algorithm was evaluated over the BRATS MICCAI 2015 dataset having 274 glioma cases, consisting both of high grade and low grade GBM. The results were compared with that of the 2D saliency detection algorithm taken over the entire sequence of brain data. For all comparisons, the Area Under the receiver operator characteristic (ROC) Curve (AUC) has been found to be more than 0.99 ± 0.01 over various tumour types, structures and locations.
Subhashis Banerjee, Sushmita Mitra, and B. Uma Shankar
IEEE
Brain tumors constitute one of the deadliest forms of cancers, with a high mortality rate. Of these, Glioblastoma multiforme (GBM) remains the most common and lethal primary brain tumor in adults. Tumor biopsy being challenging for brain tumor patients, noninvasive techniques like imaging play an important role in the process of brain cancer detection, diagnosis and prognosis; particularly using Magnetic Resonance Imaging (MRI). Therefore, development of advanced extraction and selection strategies of quantitative MRI features become necessary for noninvasively predicting and grading the tumors. In this paper we extract 56 three-dimensional quantitative MRI features, related to tumor image intensities, shape and texture, from 254 brain tumor patients. An adaptive neuro-fuzzy classifier based on linguistic hedges (ANFC-LH) is developed to simultaneously select significant features and predict the tumor grade. ANFC-LH achieves a significantly higher testing accuracy (85.83%) as compared to existing standard classifiers.
Subhashis Banerjee, Sushmita Mitra, and B. Uma Shankar
Springer Singapore
Subhashis Banerjee, Sushmita Mitra, B. Uma Shankar, and Yoichi Hayashi
Public Library of Science (PLoS)
The automatic computerized detection of regions of interest (ROI) is an important step in the process of medical image processing and analysis. The reasons are many, and include an increasing amount of available medical imaging data, existence of inter-observer and inter-scanner variability, and to improve the accuracy in automatic detection in order to assist doctors in diagnosing faster and on time. A novel algorithm, based on visual saliency, is developed here for the identification of tumor regions from MR images of the brain. The GBM saliency detection model is designed by taking cue from the concept of visual saliency in natural scenes. A visually salient region is typically rare in an image, and contains highly discriminating information, with attention getting immediately focused upon it. Although color is typically considered as the most important feature in a bottom-up saliency detection model, we circumvent this issue in the inherently gray scale MR framework. We develop a novel pseudo-coloring scheme, based on the three MRI sequences, viz. FLAIR, T2 and T1C (contrast enhanced with Gadolinium). A bottom-up strategy, based on a new pseudo-color distance and spatial distance between image patches, is defined for highlighting the salient regions in the image. This multi-channel representation of the image and saliency detection model help in automatically and quickly isolating the tumor region, for subsequent delineation, as is necessary in medical diagnosis. The effectiveness of the proposed model is evaluated on MRI of 80 subjects from the BRATS database in terms of the saliency map values. Using ground truth of the tumor regions for both high- and low- grade gliomas, the results are compared with four highly referred saliency detection models from literature. In all cases the AUC scores from the ROC analysis are found to be more than 0.999 ± 0.001 over different tumor grades, sizes and positions.