Clinical Data Analytics & Radiomics



The group’s vision and aim are development of novel clinical data analysis algorithms to extract meaningful information which enables detection, characterization, and quantification of clinically significant information. Our group aims to:

  • Develop new skills & knowledge in advanced multi-modal image (MR, CT, and PET) data processing,
  • Contribute to local research activities,
  • Forge collaborations with clinical community and industries,
  • Develop new IPs, publish in prestigious journals and secure grants.
The group’s current capabilities are in Quantitative Biology and Radiomics. 


Figure 1. Our group focuses on developing Computer aided Decision support systems for preclinical and clinical applications. The following paragraphs describe our efforts along this direction with use cases.

1. End-end DL Framework for TBI lesion segmentation on multi-contrast MR images

Traumatic brain injury (TBI) is a leading cause of death and disability. At the time of the traumatic impact, primary injury is enforced by the transfer of kinetic forces causing neurological damage. Secondary alterations of the TBI comprises of consecutive pathological processes initiated at the moment-of-injury and progressive tissue damage in the form of reduced cerebral blood flow, cerebral edema, blood-brain barrier damage, cell apoptosis and necrosis, and neuroinflammation, all of which form a very complex environment with a multitude of contrasts on MRI scans.


Figure 2. End-end DL framework for TBI lesion segmentation on multi-contrast MR images.

Areas of ongoing secondary injury are referred to as the traumatic penumbra—perilesional areas that are dynamic and constantly changing with time around the necrotic core and have the potential to recover. In the absence of timely intervention, the affected brain tissue, which could have recuperated would become an irreversible injury. Hence, accurate assessment of lesion area and its sub-regions, including penumbra’s existence is crucial for early intervention. In this study, we have developed a deep-learning framework to handle multi-contrast MR for TBI lesion and sub-region segmentation to track the pathophysiological changes in TBI for early intervention.

2. AI-augmented clinical radiology platform for detection of brain aging using Normative MRI Brain Aging Atlas, based on the local population in Singapore

Brain aging is different from chronological ageing and indicates individuals brain health. Brain age is a dynamic factor and accelerated brain ageing can be reversed using a targeted therapy. Owing to current paucity of AI-augmented advanced brain-structural changes quantifying platform, radiologists has since been using visual inspection during their routine practice. This is a subjective way of extracting information of brain atrophy/neurodegeneration and poses challenge even to the neuroradiologists.

An AI-augmented brain volumetry-based brain age detection platform (so-called ‘SaMD’) is being developed using normative population data in Singapore or ‘SG-brain’ template. This template will help to diagnose accelerated brain aging and facilitates differentiation from normal brain ageing. Many brain ageing studies were available for the Caucasian population. Owing to the difference in normative aging between the Caucasian and the Singapore population, there is a need to develop a normative atlas to detect accelerated brain aging in Singapore. It will serve as a local reference for normative brain volumetry leading to early detection, early intervention, and effective management.


Figure 3a. Data processing pipeline to generate brain atlas templates.
Figure 3b. Atlas templates of different age groups developed.

3. AI/ML based identification of Neuroanatomical markers for Schizophrenics and subtype classification

Schizophrenia is a major psychiatric disorder that imposes enormous clinical burden on patients and their caregivers. Determining classification biomarkers can complement clinical measures and improve understanding of the neural basis underlying schizophrenia. Using neuroanatomical features, several machine learning-based investigations have attempted to classify schizophrenia from healthy controls, but the range of neuroanatomical measures employed have been limited in range to date. In this study, we sought to classify schizophrenia and healthy control cohorts using a diverse set of neuroanatomical measures (cortical and subcortical volumes, cortical areas and thickness, cortical mean curvature) and adopted Ensemble methods for better performance. Additionally, we correlated such neuroanatomical features with Quality of Life (QoL) assessment scores within the schizophrenia cohort. With Ensemble methods and diverse neuroanatomical measures, we achieved classification accuracies ranging from 83 to 87%, sensitivities and specificities varying between 90–98% and 65–70% respectively. In addition to lower QoL scores within schizophrenia cohort, significant correlations were found between specific neuroanatomical measures and psychological health, social relationship subscale domains of QoL. Our results suggest the utility of inclusion of subcortical and cortical measures and Ensemble methods to achieve better classification performance and their potential impact of parsing out neurobiological correlates of quality of life in schizophrenia.

In this study, we are working on

  1. Quantitative Neuroimaging features for SCZ /HC
  2. Correlation studies

    1. Imaging features with underlying conditions
    2. Structural / Diffusion / Functional (in subtypes)
    3. Imaging features and Psychometric scores
    4. Neuroanatomy subtypes /Genetics
    5. Prediction of Remission
    6. Subtypes and Deficit conditions
  3. Neuroimaging based condition Detection/ Staging
  4. Causality b/w statistically identified imaging features with clinical symptoms.
  5. Neuroanatomy subtypes and its relation to management of the condition


Figure 4 – Neuroanatomical subtypes of Schizophrenia and their correlation to PANSS and QOL. 

4. Comprehensive Abdominal Fat Analysis Tool for Geriatric Cohort Study

Figure 5.

With the increasing prevalence of obesity and its tremendous health effects, it is important to invest time and effort into research of obesity and its associated diseases to develop better care and prevention. Worldwide, there are many clinical trials like – GENYAL (prevention of obesity in childhood); SWITCH (Community based obesity prevention trial); Clinical trial exploring exercise intervention on obese women; SAMS (Singapore adult Metabolism Study); GUSTO (Growing towards healthy outcome in Singapore); etc. and many countries like Singapore have declared war on obesity. Understanding the phenotypes and genotypes of obesity is crucial for profiling and management of the condition. Quantification of obesity in-vivo, helps monitoring the changes due to interventions, visualization of fat compartments, inter-ethnicity differences, inter – and intra-subject changes etc. which are part of the large cohort studies. In large cohort studies there is data deluge, and we need a comprehensive tool which includes (i) Accurate, rapid, and repeatable segmentation module (ii) Module for visualization of data and statistics and (iii) A correction tool to correct for errors if any.

Figure 6. End to End Human Metabolic Fat Analysis Framework.

In this study, we propose “CAFT: Comprehensive Abdominal fat segmentation tool” which comprises of a deep learning framework “Global aggregation UNet with self-attention (GA-UNet)” to accurately segment all the three compartments SSAT, DSAT and VAT simultaneously using a single network architecture. Dashboard for data presentation and analytics – which allows automatic lumbar based quantification and analysis; 3D visualization; Percentage and volumetric analysis of whole abdomen or lumbar based fat and an Interactive correction tool to edit the contours (between background & outer abdominal boundary; between SSAT & DSAT – correction of Fascia Superficialis and inner abdominal boundary which separates SAT and VAT) to correct the errors in segmentation.

5. Automated Analysis of Cardiac MRI for Clinical Trials

The proposed study aims to speed up segmentation and improve interobserver agreement in CMR imaging analysis. Automated analysis via machine learning show potential and could remove this intra- and inter-observer variation. We seek to investigate an AI algorithm for LV segmentation and function analysis to achieve robust automated results in other aspects of CMR analysis such as infarct size/T1/T2/LGE analysis. To reduce the time spent by radiologists to process data and minimize intra- and inter-observer variability, we propose a fully automated multi-scan CMR image analysis pipeline. The refined and iterated pipeline proposed aims to have equivalent precision (scan-rescan reproducibility) to a human expert with global standardization using multicentre, international, de-identified trial data from IMMACULATE/PITRI/Platelet-STEMI cohort to train further iterations. The development of this fully automated multi-scan CMR image analysis pipeline has immense translational potential and will enhance the research ecosystem in the field of ML and AI. Future directions include its application to the Singapore Biobank to develop normal and disease measurement ranges for Asian cohort, licensing to CVI42 and other image analysis vendors, provision of research CMR Core lab service with superior efficiency to international trial groups.

Background: Cardiac Magnetic Resonance (CMR) sequences are commonly used to obtain a complete description of the function and structure of the heart, provided that accurate measurements are extracted from images. New methods of extraction of information are being developed, among them, deep neural networks are powerful tools that showed the ability to perform fast and accurate segmentation. To reduce the time spent by radiologists to process data and minimize intra- and inter-observer variability, we propose a fully automatic multi-scan CMR image analysis pipeline.

Methods: Sequence specific U-Net 2D models were trained to perform the segmentation of the left ventricle (LV), right ventricle (RV) and aorta in cine short-axis, late gadolinium enhancement, sequences depending on the need. The models were trained and tested on a set of data manually segmented by experts using semi-automatic and manual tools. A set of parameters were computed from the resulting segmentations, such as the left ventricle ejection fraction (EF), scar percentage. The Dice similarity coefficient, Hausdorff distance, mean surface distance, and Pearson correlation coefficient R were used to assess and compare the results of the U-Net-based pipeline with intra-observer variability.

Figure 7. Automated analysis of cardiac MRI for clinical trials IMMAULATE/PITRI/Platelet-STEMI (With NHCS)

6. Brown Fat and Metabolic Health

Fat is not created equal. Just as there are different types of dietary fat, the human body also has different types of fat cells. Fat or adipose tissue is composed of white and brown fat – white fat stores excess energy while brown fat dissipates excess energy or fat storage as heat through a process called thermogenesis.

Brown fat or brown adipose tissue (BAT) was traditionally believed to be present abundantly only in infants and to progressively regress following infancy. Recently, studies have shown that BAT persists into adulthood. However, it’s noted that less brown fat can be found in obese individuals. BAT has since received a great deal of attention as an appealing target to combat obesity and metabolic diseases – such as diabetes – due to its ability to use body fat stores as fuel.

Boosting metabolism with brown fat

Research suggests that BAT is a major contributor to the regulation of energy metabolism in the body. Brown fat, when activated, increases thermogenesis by stimulating fatty acid oxidation and thereby increasing energy dissipation. BAT also possesses a great capacity for glucose uptake from the circulation, thereby keeping blood sugar levels balanced and improving insulin sensitivity – both of which are beneficial to an individual’s health. These properties suggest that brown fat might have an important role in whole-body metabolism.

Measuring BAT

The most commonly used method for estimating activity and quantity of brown fat in humans has been positron emission tomography (PET) combined with computed tomography (CT), which involves significant doses of radiation. Therefore, most previous studies on brown fat were done in patients requiring PET/CT scans for medical indications, limiting its applicability to the general population and to its use in children.

We investigated the influence of brown fat on adiposity in 198 young Asian preschool children (at 4.5 years of age) from the Growing Up in Singapore Towards healthy Outcomes (GUSTO) study using magnetic resonance imaging (MRI), an alternative technique to identifying brown fat without radiation. Our study is the first to examine brown fat in relation to metabolic health in a cohort of children from a general population.

The current available applied methods to identify brown fat in humans are PET/CT and infrared thermography. These techniques generally detect activated brown fat, but do not detect inactive or weakly activated brown fat. Our findings confirm that identification and quantification of brown fat is possible with MRI and strengthen the proposition of MRI as an alternative imaging technique to characterise brown fat without any form of activation. This paves a new and effective way for future studies in identifying brown fat in healthy populations, children and in longitudinal studies without applying methods involving radiation. Figure 8 explain the brown adipose tissue quantification & population template generation framework.

Figure 8. Brown adipose tissue quantification & population template generation.

Moving Forward

Figure 9. CDAR's plan moving forward


Figure 10. Collaborations


 Principal Investigator  Bhanu Prakash K.N    |    [View Bio]  
 Scientist CHILLA V.N. Geetha Soujanya 
 Senior Research Officer II YEOW Ling Yun 

Selected Publications

  • Chew, Q. H., Bhanu Prakash KN., Koh, L. Y., Chilla, G., Yeow, L. Y., & Sim, K. (2022). Neuroanatomical subtypes of schizophrenia and relationship with illness duration and deficit status. Schizophrenia Research, 248, 107-113. (Chew & Bhanu Prakash are joint first authors).

  • Geetha Soujanya Chilla, Yeow Ling Yun, Qian Hui Chew, Kang Sim, Bhanu Prakash KN. "Machine learning classification of schizophrenia patients and healthy controls using diverse neuroanatomical markers and Ensemble methods.Sci Rep 12, 2755 (2022). [IF: 5.035]  

  • Eelin Tan; Khurshid Merchant; Bhanu Prakash KN; Arvind Channarayapatna Srinivasa; et al. (2022).CT-based Morphologic and Radiomics Features for the Classification of MYCN Gene Amplification Status in Pediatric Neuroblastoma”. Childs Nerv Syst. 2022 Apr 23. doi: 10.1007/s00381-022-05534-3. Epub ahead of print. PMID: 35460355. [IF: 1.48]

  • Mazira Mohammad Ghazali, Che Mohd Nasril Che Mohd Nassir, Nur Suhaila Idris, Geetha Chilla, Bhanu Prakash KN, Muzaimi Mustapha.Presence of Enlarged Perivascular Spaces Is Associated with Reduced Processing Speed in Apparently Asymptomatic, Working-Aged Adults”. J Integr Neurosci. 2022 Mar 21;21(2):51. doi: 10.31083/j.jin2102051. PMID: 35364639. [IF: 2.117]

  • Bhanu Prakash KN, Arvind CS, Yeow LY, Chen WX, Lim WS, Tan CH. CAFT: A Deep Learning based Comprehensive Abdominal Fat Analysis Tool for Large Cohort Studies. Magnetic Resonance Materials in Physics, Biology and Medicine, 1-16. 2 Aug 2021 DOI: [IF: 2.310] 

  • Arvind CS, Ling Yun Yeow, Chen Wen Xiang, Lim Wee Shiong, Cher Heng Tan, Bhanu Prakash KN (2021),Residual Global Attention U-Net Based Fat Depot Segmentation for Sarcopenic Obesity Profiling” SGCR-WIRES.

  • Bhanu Prakash KN, Arvind CS, Ling Yun Yeow, Chen Wen Xiang, Audrey Yeo, Wee Shiong Lim, Cher Heng Tan. (2021). Differential association of abdominal fat depot measures on MRI with body composition phenotypes and anthropometric measures. 8th Asian SGCR- WIRES Conference Aug 11-15, Singapore.

  • Bhanu Prakash KN, Arvind CS, Ling Yun Yeow, Chen Wen Xiang, Audrey Yeo, Wee Shiong Lim, Cher Heng Tan. AI-based Clinical assessment framework for SG Phenotype Sarcopenic obesity profiling and differential association with anthropometric measures. A*STAR Scientific Conference. 2021.

  • Che Mohd Nassir CMN, Damodaran T, Yusof SR, Norazit A, Chilla G, Huen I, Bhanu Prakash KN, Mohamed Ibrahim N, Mustapha M. Aberrant Neurogliovascular Unit Dynamics in Cerebral Small Vessel Disease: A Rheological Clue to Vascular Parkinsonism. Pharmaceutics 2021, 13, 1207.

  • Ren BX, Huen I, Wu ZJ, Wang H, Duan MY, Guenther I, Bhanu Prakash KN, Tang FR. Early Postnatal Irradiation-induced Age-dependent Changes in Adult Mouse Brain: MRI based Characterization. BMC Neuroscience, 22, Article number: 28 (2021). 

  • Gawali M , Arvind CS, Suryavanshi S, Madaan H, Gaikwad A, Bhanu Prakash KN, Kulkarni V, Pant A. Comparison of Privacy-Preserving Distributed Deep Learning Methods in Healthcare. arXiv:2012.12591, 2021. 

  • Manish Gawali, Arvind C S, Shriya Suryavanshi, Harshit Madaan, Ashrika Gaikwad, Bhanu Prakash KN, Viraj Kulkarni, Aniruddha Pant.(2021).Comparison of Privacy-Preserving Distributed Deep Learning Methods in Healthcare”. In: Papież B.W., Yaqub M., Jiao J., Namburete A.I.L., Noble J.A. (eds) Medical Image Understanding and Analysis. MIUA 2021. Lecture Notes in Computer Science, vol 12722. Springer, Cham.

  • View full list of publications and patents here