Deep Learning-Based Diagnostics Of Head And Neck Cancers Using Coherent Raman Scattering Microscopy

Bjarne Thorsted

Research output: ThesisPh.D. thesis

Abstract

head and neck cancers (HNCs) are rare cancers of mainly squamous cell carcinoma
(SCC) that do not show symptoms typically associated with cancer in
the early stages of development. The main causes are alcohol and tobacco
consumption with human papillomavirus (HPV) also becoming an increasingly
common cause. While treatment is uncomplicated in the early stages,
mortality significantly increases if patients do not seek medical attention
before the cancer has developed into the later stages. Treatment typically
involves surgery and chemoradiotherapy, where the former carries the risk
of damaging the delicate organs in the head and neck region. The interface
between healthy tissue and tumor is subtle and gradual, which means that
surgeons are often dependent on intra-operative consultations (IOCs) during
operation to determine whether enough tissue has been resected. This process
involves transferring (a part of) the resected tissue to the Department of
Pathology for macroscopic inspection, flash freezing, microsectioning and
microscopic assessment before a response can be returned to the surgeon
waiting in the operating theater. Reducing this lag time would result in
cost reductions and potentially increased patient throughput as well as the
workload at the Department of Pathology.
In this thesis the applicability of a deep neural network (DNN) as an
automated diagnosing tool was investigated. The hypothesis being that noninvasive
imaging modalities, specifically coherent anti-Stokes Raman scattering
(CARS), second-harmonic generation (SHG), and two-photon excited fluorescence
(TPEF) microscopy, could provide adequate information about a
label-free sample for a DNN to learn to recognize different tissue types if
presented with a database of such images.
The study in this project shows that the combination of CARS, SHG, and
TPEF deliver enough information to positively identify several important
features of both healthy, dysplasic, and cancerous tissue. Furthermore, it
also showed that a fully convolutional neural network (FCNN) can be trained
from a relatively small image database and achieve an impressive accuracy.
Several hyper parameters were evaluated and we found that the highest
performance was achieved with a combination of the novel Swish activation
function, batch normalization, the region-based Tversky loss function and a
host of input data augmentation algorithms including, but not limited to,
random pixel deletion (dropout), rotation and elastic deformation.
The results of this study shows that multi-modal imaging in combination
with deep learning-based analysis in the future can help reduce the pathologists
workload during IOCs and that more data is needed to improve the
performance.
Original languageEnglish
Awarding Institution
  • University of Southern Denmark
Supervisors/Advisors
  • Brewer, Jonathan, Supervisor
  • Godballe, Christian, Supervisor
  • Larsen, Stine Rosenkilde, Supervisor, External person
Publisher
Publication statusPublished - Oct 2020

Fingerprint

Dive into the research topics of 'Deep Learning-Based Diagnostics Of Head And Neck Cancers Using Coherent Raman Scattering Microscopy'. Together they form a unique fingerprint.

Cite this