BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:US/Eastern BEGIN:DAYLIGHT DTSTART:20230312T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20231105T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20230825T230242Z UID:0DF23FA3-F089-4864-9DC5-D0AE8E7AB68C DTSTART;TZID=US/Eastern:20230825T180000 DTEND;TZID=US/Eastern:20230825T190000 DESCRIPTION:The [IEEE Long Island (LI) Signal Processing Society (SPS)](htt ps://ieee.li/society-chapters/signal-processing-society-sp/) in collaborat ion with [North Jersey Social Implications of Technology Society](https:// r1.ieee.org/northjersey/chapter/ssit/) presents the following Technical Le cture:\n\nAccurate image segmentation holds significance for vital clinica l applications such as diagnosis and surgery planning. While deep neural n etworks have excelled in achieving superior segmentation outcomes via full y supervised learning\, their reliance on substantial annotated training d ata is a challenge. Procuring extensive labeled datasets for medical image s is labor-intensive and costly due to the need for clinical expertise in annotations. Thus\, an opportunity for improvement is evident. Hence\, the critical need to devise strategies for attaining medical images with scan t annotations while harnessing untapped potential within unlabeled data du ring training. We harness the power of self-supervised representation lear ning and semi-supervised learning in this regard and perform extensive exp eriments on images from multiple modalities: Computer Tomograhpy (CT) scan \, Magnetic Resonance Imaging (MRI) scan\, Histopathology studies\, etc. O ur recent research showcases that even with minimal annotations estimate o f x<10%\, we achieve comparable or superior performance compared to fully supervised approaches.\n\nSpeaker(s): Mr. Hritam Basak\, Student\, \n\nAge nda: \nTechnical support set-up: 5:30pm EST\nIntroductions 6pm-6:05pm EST\ nTechnical Lecture: 6:05pm-6:50pm EST\nQ&A: 6:50pm-7pm EST\n\nVirtual: htt ps://events.vtools.ieee.org/m/369575 LOCATION:Virtual: https://events.vtools.ieee.org/m/369575 ORGANIZER:Signal@ieee.li SEQUENCE:99 SUMMARY:Maximizing Learning with Minimal Labels: Innovations in Medical Ima ge Analysis with Sparse Labels URL;VALUE=URI:https://events.vtools.ieee.org/m/369575 X-ALT-DESC:Description: <br /><p style="font-weight: 400\;">The <a href="ht tps://ieee.li/society-chapters/signal-processing-society-sp/">IEEE Long Is land (LI) Signal Processing Society (SPS)</a> in collaboration with <a hre f="https://r1.ieee.org/northjersey/chapter/ssit/">North Jersey Social Impl ications of Technology Society</a> presents the following Technical Lectur e:</p>\n<p style="font-weight: 400\;">Accurate image segmentation holds si gnificance for vital clinical applications such as diagnosis and surgery p lanning. While deep neural networks have excelled in achieving superior se gmentation outcomes via fully supervised learning\, their reliance on subs tantial annotated training data is a challenge. Procuring extensive labele d datasets for medical images is labor-intensive and costly due to the nee d for clinical expertise in annotations. Thus\, an opportunity for improve ment is evident. Hence\, the critical need to devise strategies for attain ing medical images with scant annotations while harnessing untapped potent ial within unlabeled data during training. We harness the power of self-su pervised representation learning and semi-supervised learning in this rega rd and perform extensive experiments on images from multiple modalities: & nbsp\;Computer Tomograhpy (CT) scan\, Magnetic Resonance Imaging (MRI) sca n\, Histopathology studies\, etc. Our recent research showcases that even with minimal annotations estimate of x&lt\;10%\, we achieve comparable or superior performance compared to fully supervised approaches.</p><br /><br />Agenda: <br /><p>Technical support set-up: 5:30pm EST<br />Introduction s 6pm-6:05pm EST<br />Technical Lecture: 6:05pm-6:50pm EST<br />Q&amp\;A: 6:50pm-7pm EST</p> END:VEVENT END:VCALENDAR