Computational Modelling of Tissue Energy Interaction in Acoustic and Optical Imaging for In situ Diagnostic Histopathology
by Debdoot Sheet
Submitted to Indian Institute of Technology Kharagpur in Fulfillment of the Requirements for the Award of the Degree of
Doctor of Philosophy
Abstract
S
OFT TISSUESincluding blood vessels get affected by wounds, abnormal deposition of extracellular matrix, hyperplasia or hypoplasia in epithelia, as well as growth of benign and malignant tumours which are medically termed as lesions. In diagnostics, a small quantity of tissue components are col- lected from the lesion by biopsy or through aspiration, followed by cyto-/histological processing and interpretation by an expert Pathologist. Such invasive procedure causes patient discomfort and results in about48−72hours of delay in reporting. Moreover, this practice is not feasible for investigating crit- ical organs, vessels and healing wound beds. Although, subsurface imaging like ultrasonography (US) and optical coherence tomography (OCT) provide option ofin situinvestigation, the clinically relevant information yield is much less than conventional histopathology.This thesis presents computational models for processing of acoustic and optical subsurface imaging signals and evaluates its efficacy forin situdiagnostic histopathology of soft tissue in real time. The rational was to develop algorithms for solving a set of statistical physics equations to model tissue energy interaction in acoustic imaging with US and optical imaging with white light ophthalmoscopy and OCT.
In acoustic imaging ultrasonic propagation and backscattering in heterogeneous tissues are mod- elled and used to characterize tissues. In intravascular ultrasound (IVUS), a model for uncompressed signals identifies fibrous tissue, calcified, lipidic plaque and necrosis in atherosclerosis at sensitivity of 97%,99%,99%and96%respectively with specificity above80%. In B-mode ultrasonic imaging, the model identifies BIRADS Category 2,3,4,5 breast lesions at100%,100%,100%and95%detection rates respectively with classification rates as area under ROC of0.99,0.99,0.98and0.87.
Retinal vessels are detected by modelling the statistical physics of non-ballistic photon imaging for white light fundus angiography with an accuracy of98% and ground truth consensus at kappa score of 0.83. Skin is characterized by modelling of ballistic and near-ballistic photon imaging in swept source OCT to characterize epidermis, papillary dermis, dermis and adipose tissue in healthy and wound tissue at sensitivity of99%,95%,99%and99%. This work comprehensively addresses the challenges in subsurface acoustic and optical imaging and presents patient comfort centricin situhistopathology solutions by learning the stochastic uncertainty of tissue energy interaction.
Keywords: In situhistopathology, transfer learning, ultrasonic backscattering statistics, signal attenua- tion, intravascular ultrasound, tissue photon interaction, ophthalmoscopy, optical coherence tomography.
Supervisors: Professor Ajoy Kumar Ray (Professor of Electronics and Electrical Communication En- gineering) and Dr. Jyotirmoy Chatterjee (Associate Professor of Medical Science and Technology)