I am currently a Senior Research Scientist at Siemens Healthineers Princeton NJ.

abstract | pdf | bibtex
IEEE Journal of Biomedical and Health Informatics, J-BHI, 2020
Impact Factor: 5.180

CNN based lung segmentation models in absence of diverse training dataset fail to segment lung volumes in presence of severe pathologies such as large masses, scars, and tumors. To rectify this problem, we propose a multi-stage algorithm for lung volume segmentation from CT scans. The algorithm uses a 3D CNN in the first stage to obtain a coarse segmentation of the left and right lungs. In the second stage, shape correction is performed on the segmentation mask using a 3D structure correction CNN. A novel data augmentation strategy is adopted to train a 3D CNN which helps in incorporating global shape prior. Finally, the shape corrected segmentation mask is up-sampled and refined using a parallel flood-fill operation. The proposed multi-stage algorithm is robust in the presence of large nodules/tumors and does not require labeled segmentation masks for entire pathological lung volume for training. Through extensive experiments conducted on publicly available datasets such as NSCLC, LUNA, and LOLA11 we demonstrate that the proposed approach improves the recall of large juxtapleural tumor voxels by at least 15% over state-of-the-art models without sacrificing segmentation accuracy in case of normal lungs. The proposed method also meets the requirement of CAD software by performing segmentation within 5 seconds which is significantly faster than present methods..

@article{9122557,
  author={P. {Sahu} and Y. {Zhao} and P. {Bhatia} 
  and L. {Bogoni} and A. {Jerebko} and H. {Qin}},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={Structure Correction for Robust Volume 
  Segmentation in Presence of Tumors}, 
  year={2020},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/JBHI.2020.3004296}}

abstract | pdf | bibtex
IEEE Journal of Biomedical and Health Informatics, J-BHI, 2018
Impact Factor: 5.180

The size and shape of a nodule are the essential indicators of malignancy in lung cancer diagnosis. However, effectively capturing the nodule\92s structural information from CT scans in a Computer-aided system is a challenging task. Unlike previous models which proposed computationally intensive deep ensemble models or 3D CNN models, we propose a lightweight, multiple view sampling based Multi-section CNN architecture. The model obtains a nodule\92s cross-sections from multiple view angles and encodes the nodule\92s volumetric information into a compact representation by aggregating information from its different cross-sections via a view pooling layer. The compact feature is subsequently used for the task of nodule classification. The method does not require nodule\92s spatial annotation and works directly on the crosssections generated from volume enclosing the nodule. We evaluated the proposed method on LIDC-IDRI dataset. It achieved state-of-the-art performance with a mean 93.18% classification accuracy. The architecture could also be used to select the representative cross-sections determining nodule\92s malignancy which facilitates in the interpretation of results. Because of being lightweight the model could be ported to mobile devices which brings the power of AI driven application directly into practitioner\92s hand.

@article{sahu2018lightweight,
  title={A lightweight multi-section CNN for lung nodule 
  classification and malignancy estimation},
  author={Sahu, Pranjal and Yu, Dantong and Dasari, 
  Mallesham and Hou, Fei and Qin, Hong},
  journal={IEEE journal of biomedical and health informatics},
  volume={23},
  number={3},
  pages={960--968},
  year={2018},
  publisher={IEEE}
}

abstract | pdf | bibtex
The British Machine Vision Conference, BMVC, 2021

We propose a novel stabilized semi-supervised training method to solve the challenging problem of covid lesion segmentation in CT scans. We first study the limitations of current models and based on our findings we introduce a lightweight SU-Net (Small U-Net) architecture. During training we feed the CT scans in sorted order of lesion occupancy and calculate a reliability score at each epoch to determine the stopping criteria. We test the proposed method on the largest publicly available COVID CT dataset called MOSMED dataset. By harnessing around 800 un-labelled COVID CT volumes comprising 25k CT slices, we improve the segmentation accuracy by around 2-4 dice percentage points depending upon the availability of labelled training data. We also compare our method with a recently published COVID lesion segmentation method called Semi-InfNet. The proposed method outperforms Semi-InfNet model and achieves state-of-the-art covid segmentation result on MOSMED dataset.

@inproceedings{sahu2021interactive,
  title={Interactive Smoothing Parameter Optimization in DBT Reconstruction Using Deep Learning},
  author={Sahu, Pranjal and Kumar, Vunnava Saikiran and Qin, Hong},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year={2021}
}

      

abstract | pdf | bibtex
Medical Image Computing and Computer Assisted Intervention, MICCAI, 2021

Medical image reconstruction algorithms such as PenalizedWeighted Least Squares (PWLS) typically rely on a good choice of tuningparameters such as the number of iterations, the strength of regularizar,etc. However, obtaining a good estimate of such parameters is often doneusing trial and error methods. This process is very time consuming andlaborious especially for high resolution images. To solve this problemwe propose an interactive framework. We focus on the regularizationparameter and train a CNN to imitate its impact on image for varyingvalues. The trained CNN can be used by a human practitioner to tunethe regularization strength on-the-fly as per the requirements. Taking theexample of Digital Breast Tomosynthesis reconstruction, we demonstrate the feasibility of our approach and also discuss the future applications of this interactive reconstruction approach. We also test the proposed methodology on public Walnut and Lodopab CT reconstruction datasetsto show it can be generalized to CT reconstruction as well.

@inproceedings{sahu2021interactive,
  title={Interactive Smoothing Parameter Optimization in DBT Reconstruction Using Deep Learning},
  author={Sahu, Pranjal and Huang, Hailiang and Zhao, Wei and Qin, Hong},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={57--67},
  year={2021},
  organization={Springer}
}

      

abstract | pdf | bibtex
INFOCOM, 2020

360 degree videos provide an immersive experience to users, but require considerably more bandwidth to stream compared to regular videos. State-of-the-art 360 \BF video streaming systems use viewport prediction to reduce bandwidth requirement , that involves predicting which part of the video the user will view and only fetching that content. However, viewport prediction is error prone resulting in poor user Quality of Experience (QoE). We design PARSEC, a 360 \BF video streaming system that reduces bandwidth requirement while improving video quality. PARSEC trades off bandwidth for additional client-side computation to achieve its goals. PARSEC uses an approach based on super-resolution, where the video is significantly compressed at the server and the client runs a deep learning model to enhance the video to a much higher quality. PARSEC addresses a set of challenges associated with using super-resolution for 360 \BF video streaming: large deep learning models, slow inference rate, and variance in the quality of the enhanced videos. To this end, PAR-SEC trains small micro-models over shorter video segments, and then combines traditional video encoding with super-resolution techniques to overcome the challenges. We evaluate PARSEC on a real WiFi network, over a broadband network trace released by FCC, and over a 4G/LTE network trace. PARSEC significantly outperforms the state-of-art 360 \BF video streaming systems while reducing the bandwidth requirement.

@inproceedings{dasari2020streaming,
  title={Streaming 360-degree videos using super-resolution},
  author={Dasari, Mallesham and Bhattacharya, Arani and Vargas, Santiago and Sahu, Pranjal and 
  Balasubramanian, Aruna and Das, Samir R},
  booktitle={IEEE INFOCOM 2020-IEEE Conference on Computer Communications},
  pages={1977--1986},
  year={2020},
  organization={IEEE}
}

abstract | pdf | bibtex
IWBI, 2020 (Oral)

Contrast enhanced digital breast tomosynthesis (CEDBT) utilizes weighted subtraction of high energy (HE) and low energy (LE) DBT to generate a 3D iodinated contrast enhancement map of the breast, and potentially improve breast lesion detection and characterization. However, the increased scattered radiation at HE exacerbates the cupping artifact. Monte Carlo (MC) based scatter correction (SC) method suffers from long computation time, and kernel-based method is less accurate, especially near the breast edge due to thickness roll-off. This work is aimed at developing fast and accurate SC using Convolutional Neural Network (CNN).

@inproceedings{duan2020scatter,
  title={Scatter correction with deep learning
  approach for contrast enhanced digital
  breast tomosynthesis (CEDBT) in both
  cranio-caudal (CC) view and mediolateral oblique (MLO) view},
  author={Duan, Xiaoyu and Sahu, Pranjal and
  Huang, Hailiang and Zhao, Wei},
  booktitle={15th International Workshop
  on Breast Imaging (IWBI2020)},
  volume={11513},
  pages={115130Q},
  year={2020},
  organization={International Society
  for Optics and Photonics}
}

abstract | pdf | bibtex
IEEE International Symposium on Biomedical Imaging, ISBI, 2019

Digital Breast Tomosynthesis (DBT) provides a quasi-3D impression of the breast volume resulting in a better visualization of mass. However, one serious drawback of Tomosynthesis is that compared to Mammography, each projection gets lower x-ray dose resulting into higher quantum noise which seriously hampers the visibility of calcifications. To solve this problem we propose a Convolutional Neural Network model based on Adversarial loss. We train the deep network using synthetic data obtained from Virtual Clinical Trials. Unlike earlier works which tested model on phantoms only, we performed experiments on real samples obtained in clinical settings as well. Our approach shows encouraging results in denoising the projections. De-noised projections show higher perceptual similarity with mammograms and superior signalto- noise ratio. The reconstructed volume also enhances calcification visibility. Our work shows the viability of utilizing synthetic data for training the deep network for de-noising purposes.

@inproceedings{sahu2019using,
  title={Using Virtual Digital Breast Tomosynthesis 
  for De-Noising of Low-Dose Projection Images},
  author={Sahu, Pranjal and Huang, Hailiang and 
  Zhao, Wei and Qin, Hong},
  booktitle={2019 IEEE 16th International Symposium on 
  Biomedical Imaging (ISBI 2019)},
  pages={1647--1651},
  year={2019},
  organization={IEEE}
}

abstract | pdf | bibtex
Medical Imaging, SPIE, 2018 (Best Demo Award)

Melanoma is the most dangerous form of skin cancer that often resembles moles. Dermatologists often rec-ommend regular skin examination to identify and eliminate Melanoma in its early stages. To facilitate thisprocess, we propose a hand-held computer (smart-phone, Raspberry Pi) based assistant that classifies with thedermatologist-level accuracy skin lesion images into malignant and benign and works in a standalone mobiledevice without requiring network connectivity. In this paper, we propose and implement a hybrid approachbased on advanced deep learning model and domain-specific knowledge and features that dermatologists usefor the inspection purpose to improve the accuracy of classification between benign and malignant skin lesions.Here, domain-specific features include the texture of the lesion boundary, the symmetry of the mole, and theboundary characteristics of the region of interest. We also obtain standard deep features from a pre-trainednetwork optimized for mobile devices called Google\92s MobileNet. The experiments conducted on ISIC 2017 skincancer classification challenge demonstrate the effectiveness and complementary nature of these hybrid featuresover the standard deep features. We performed experiments with the training, testing and validation data splitsprovided in the competition. Our method achieved area of 0.805 under the receiver operating characteristiccurve. Our ultimate goal is to extend the trained model in a commercial hand-held mobile and sensor devicesuch as Raspberry Pi and democratize the access to preventive health care.

  @inproceedings{sahu2018apply,
  title={Apply lightweight deep learning 
  on internet 
  of things for low-cost and 
  easy-to-access 
  skin cancer detection},
  author={Sahu, Pranjal and Yu, 
  Dantong and Qin, Hong},
  booktitle={Medical Imaging 2018: 
  Imaging Informatics 
  for Healthcare, Research, and Applications},
  volume={10579},
  pages={1057912},
  year={2018},
  organization={International Society 
  for Optics and Photonics}
}