BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:85@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20241212T140000
DTEND;TZID=Asia/Kolkata:20241212T150000
DTSTAMP:20241212T055632Z
URL:https://cds.iisc.ac.in/events/m-tech-research-thesis-colloquium-cds-le
 sion-synthesis-using-physics-based-noise-models-for-low-data-medical-imagi
 ng-regime-applications/
SUMMARY:Re-scheduled: M.Tech Research Thesis {Colloquium}: CDS: "Lesion Syn
 thesis using Physics-Based Noise Models for Low-Data Medical Imaging Regim
 e applications"
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\n\nM.Tech Researc
 h Thesis Colloquium\n\n\n\nSpeaker : Mr. Ramanujam Narayanan\n\nS.R. Numbe
 r : 06-18-01-10-22-22-1-21772\n\nTitle : "Lesion Synthesis using Physics-B
 ased Noise Models for Low-Data Medical Imaging Regime applications"\n\nRes
 earch Supervisor : Dr. Vaanathi Sundaresan\n\nDate &amp\; Time : December 
 11\, 2024\, 11.00 am\n\nVenue : # 102 CDS Seminar Hall\n\n\n\nABSTRACT\n\n
 Lesion segmentation and their progression prediction in medical imaging re
 lies critically on the availability of manually annotated\, heterogeneous 
 large pathological datasets. Acquiring such diverse large datasets is also
  challenging because they require coordination and ethical clearances from
  multiple sites\, and the manual annotation of these images is both time-c
 onsuming and expensive. On the other hand\, this data diversity is difficu
 lt to achieve in low-data regimes\, hindering the robust training of the m
 odels.\n\nOur study presents a lesion simulation method involving structur
 al localized perturbation of healthy tissue using noise models based on th
 e physics of modalities. Later\, we localize these perturbations within ma
 sks defined by composites of ellipsoidal polygons (thus forming random sha
 pes) and blended them with the input image with varying contrast. The lesi
 on simulation step does not require training and can generate any number o
 f lesions with texture\, size\, and scale variations\, injecting sufficien
 t variability in the training data pool in low-data regimes. We evaluate t
 he performance of our simulated lesions for a downstream lesion segmentati
 on task and show superior performance than its fully supervised counterpar
 t. We also performed extensive ablation studies and experimentally determi
 ned the optimal simulation data and minimal training data required for tra
 ining the segmentation model. Further\, we also showed detailed variation 
 charts depicting the possible simulations the method can generate across d
 ifferent datasets. We evaluate the performance on publicly available patho
 logical brain MRI\, liver CT\, retinal fundus imaging and breast Ultrasoun
 d datasets with diverse lesions. Using only 75% of labelled real-world dat
 a\, the proposed method significantly improves the segmentation performanc
 e compared to the fully supervised training\, with a 16% mean increase in 
 the Dice score (DSC) and a 33% mean decrease in the 95th percentile of the
  normalised Hausdorff distance (HD95 (norm)).\n\nWe also discuss two poten
 tial use cases of the method: (i) for improving segmentation of lacune inf
 arcts by providing simulated instances for augmenting with real lacunar in
 farct dataset to training lacune segmentation model with improved performa
 nce\, (ii) for improving the post-operation (post-op) prognosis prediction
  of glioblastomas using Response Assessment in Neuro-Oncology (RANO) crite
 rion-informed simulations augmenting real post-op and follow-up scans.\n\n
 \n\nALL ARE WELCOME
CATEGORIES:Events,MTech Research Thesis Colloquium
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20231213T140000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR