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:12@cds.iisc.ac.in DTSTART;TZID=Asia/Kolkata:20230911T100000 DTEND;TZID=Asia/Kolkata:20230911T110000 DTSTAMP:20231102T051329Z URL:https://cds.iisc.ac.in/events/ph-d-thesis-colloquium-cds-unsupervised- test-time-adaptation-for-patient-specific-deep-learning-models-in-medical- imaging/ SUMMARY:Ph.D. Thesis {Colloquium}: CDS : “Unsupervised test-time adaptati on for patient-specific deep learning models in medical imaging.” DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Col loquium\n\n_______________________________________________________________ ___________________________\n\nSpeaker : Mr. Hariharan Ravishankar\n\nS.R. Number : 06-18-00-11-12-20-1-18917\n\nTitle : “Unsupervised test-time a daptation for patient-specific deep learning models in medical imaging”\ n\nResearch Supervisor: Prof. Phaneendra Kumar Yalavarthy\n\nDate &\; T ime : September 11\, 2023 (Monday) at 10:00 AM\n\nVenue : # 102 CDS Semina r Hall\n\n________________________________________________________________ __________________________\nAbstract\nDeep learning (DL) models have achie ved state-of-the-art results in multiple medical imaging applications\, re sulting in the widespread adoption of artificial intelligence (AI) models for radiological workflows. Despite their success\, they exhibit two major weaknesses in real-world applications: a) poor generalization and b) lack of patient optimality. The potential “distribution shift” in unseen m edical imaging data\, owing to changes in demography\, acquisition hardwar e\, etc.\, often leads to a reduction in performance. Despite the high ave rage performance\, they are prone to failures on “individual” cases wi th minor input modulations than the training data. Solving this problem is critical in medical image analysis because this variance in performance a cross cases and abject failures on select subjects will increase the burde n on care-giving experts and reduce trust in AI-based applications.\n\nThe aim of this thesis work is to develop methods for patient-specific test-t ime adaptation of pre-trained DL models on individual subject data. While generalization is often treated as a training time goal\, test-time adapta tion (TTA) modifies the weights during inference time in an unsupervised m anner specific to each subject data to jointly address generalization and patient optimality. While sustaining performance in individual subjects is seen as a challenge for current DL models\, this line of research also ha s an opportunity for next-generation clinical decision support systems. AI methods that would result in patient-specific DL models can be heralded a s the next big step in medical imaging.\n\nInformation Geometric Test Time Adaptation (IGTTA) for semantic segmentation: The test-time adaptation (T TA) of deep-learning-based semantic segmentation models\, specific to indi vidual patient data\, was addressed in this part of thesis work. Existing TTA methods in medical imaging are often unconstrained and require prior a natomical information or additional neural networks built during the train ing phase\, making them less practical and prone to performance deteriorat ion. In this part of thesis work\, a novel framework based on information geometric principles was proposed to achieve generic\, off-the-shelf\, reg ularized patient-specific adaptation of models during test-time. By consid ering the pre-trained model and adapted models as part of statistical neur omanifolds\, test-time adaptation was treated as constrained functional re gularization using information geometric measures\, leading to improved ge neralization and patient optimality. The efficacy of the proposed approach was shown on three challenging problems: a) improving the generalization of state-of-the-art models for segmenting COVID-19 anomalies in Computed T omography (CT) images\, b) cross-institutional brain tumor segmentation fr om magnetic resonance (MR) images\, and c) segmentation of retinal layers in Optical Coherence Tomography (OCT) images. Furthermore\, it was demonst rated that robust patient-specific adaptation can be achieved without addi ng a significant computational burden\, making it the first of its kind ba sed on information geometric principles.\n\nTest-time adaptation via domai n transforms (TTADT): OCT imaging has emerged as the modality of choice fo r the diagnosis of retinal diseases. The popular\, cost-effective variant of OCT – Spectral-Domain Optical Coherence Tomography (SDOCT) often suff ers in image quality owing to speckle noise\, making expert or automated a nalysis of these images challenging. This part of thesis work proposes for the first-time inference time adaptation of deep learning models for a gi ven test sample to achieve improved diagnostic performance. The adaptation is driven by domain transforms – a framework in which domain-specific a ugmentation is employed for a test sample\, and the deep learning model we ights are updated in a constrained and regularized manner. Performance of the proposed approach was evaluated on both simulated and real-world noisy B-mode OCT images affected by varying degrees of speckle noise from four benchmarking SD-OCT datasets. Systematic studies in this part of thesis wo rk clearly demonstrate the utility of the proposed approach in building ge neralized and robust deep learning models for automated retinal disease di agnosis using OCT images.\n\nUncertainty-aware test-time adaptation for in verse problems (UATTA): Deep learning models for inverse problems (both on e-pass and unrolled versions): have these limitations 1) inability to prov ide pixel-wise uncertainty quantification 2) sensitivity to variation in f orward model and 3) sub-optimal regularization over different iterations. In this part of thesis work\, uncertainty aware test-time adaptation of of f-the-shelf models is proposed for improved image reconstruction performan ce. Different pixel-wise uncertainty quantification methods were explored for utility in test-time adaptation\, along with data fidelity constraints . The results were studied for two relevant problems:1) accelerated MRI ac quisition and 2) quantitative susceptibility mapping for quantifying tissu e magnetic susceptibility. The proposed framework provides a reliable and accurate reconstruction of these images.\n\nTest-time adaptation with foun dational models as neural surrogates (TTANS): Recently\, foundational mode ls have demonstrated impressive promptable segmentation of natural images. However\, these models often perform poorly with medical imaging data. In this part of the thesis\, a two-way synergic approach for jointly utilizi ng foundational models while adapting pretrained medical imaging segmentat ion models is explored. The results are demonstrated on various medical im aging semantic segmentation tasks to show the ability of foundational mode ls as neural surrogates\, thus demonstrating test-time adaptation achieved in real time. As these foundational models are capable of handling high l evels of semantics\, using them as surrogates helped reduce reliance on co stly annotated data needed to achieve state-of-the-art results.\n\nIn summ ary\, this thesis work developed methods to improve the generalization of typical deep-learning models utilized in medical imaging by personalizing them to individual patients during testing in an unsupervised and automate d manner. The developed methods demonstrated the applicability and utility in different medical imaging tasks across various medical imaging modalit ies. This thesis work is the first of its kind in medical imaging to showc ase the utility of test-time adaptation methods\, paving the way for furth er research towards the goal of precision medicine.\n\n=================== ============================================================\n\nALL ARE WE LCOME CATEGORIES:Events,Ph.D. Thesis Colloquium END:VEVENT BEGIN:VTIMEZONE TZID:Asia/Kolkata X-LIC-LOCATION:Asia/Kolkata BEGIN:STANDARD DTSTART:20220911T100000 TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE END:VCALENDAR