Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

A Machine Learning Approach to Skin Cancer Delineation on Photoacoustic Imaging

Fracchia, Alice LU (2023) FYTM05 20231
Department of Physics
Computational Biology and Biological Physics - Undergoing reorganization
Abstract
Skin cancer is a growing public health concern due to its prevalence among the population. Current clinical procedures require high invasiveness and multiple surgeries, which are responsible for patient discomfort and high medical expenses. Photoacoustic imaging offers an alternative to standard skin carcinoma diagnosis by exploiting the low scattering rate of ultrasound, which enables deep tissue penetration, and high imaging resolution. This thesis focuses on the application of machine learning models to systematically identify and delineate tumour regions according to their photoacoustic spectra. This is achieved through the implementation of a multilayer perceptron and a convolutional neural network that binary classify the spectral... (More)
Skin cancer is a growing public health concern due to its prevalence among the population. Current clinical procedures require high invasiveness and multiple surgeries, which are responsible for patient discomfort and high medical expenses. Photoacoustic imaging offers an alternative to standard skin carcinoma diagnosis by exploiting the low scattering rate of ultrasound, which enables deep tissue penetration, and high imaging resolution. This thesis focuses on the application of machine learning models to systematically identify and delineate tumour regions according to their photoacoustic spectra. This is achieved through the implementation of a multilayer perceptron and a convolutional neural network that binary classify the spectral inputs. The same procedure is repeated on data that were dimensionally reduced by an autoencoder. The model predictions are further refined through post-processing contouring techniques. We apply our approach to six samples of the most common skin tumour types and successfully estimate the carcinoma extension. Specifically, the convolutional network accurately estimated the tumour extension, this way consistently
decreasing the size of excision margins. This model combined with the contouring technique constitutes a safe approach to skin cancer diagnosis. Our method significantly reduces the number of required surgeries and once automated will decrease the current medical staff workload. (Less)
Popular Abstract
Skin carcinoma is one of the most common types of cancer and is on the rise among the Caucasian population. When not treated in time, this tumour is fatal. Even though several methods to diagnose skin carcinoma exist, most are limited by invasiveness or poor resolution at increasing depths, which is essential when applied to our bodies. Removal of the suspected tissue and microscopic examination is the most widely used technique. The tumour area is extracted with almost a 1-centimetre margin to ensure that all the carcinogenic regions are removed. This is a lot when the average size of a tumour is around a few millimetres. Unfortunately, sometimes this margin is still not enough, and another excision needs to be made.

Newly emergent... (More)
Skin carcinoma is one of the most common types of cancer and is on the rise among the Caucasian population. When not treated in time, this tumour is fatal. Even though several methods to diagnose skin carcinoma exist, most are limited by invasiveness or poor resolution at increasing depths, which is essential when applied to our bodies. Removal of the suspected tissue and microscopic examination is the most widely used technique. The tumour area is extracted with almost a 1-centimetre margin to ensure that all the carcinogenic regions are removed. This is a lot when the average size of a tumour is around a few millimetres. Unfortunately, sometimes this margin is still not enough, and another excision needs to be made.

Newly emergent imaging methods based on the propagation of high-frequency sound in our body offer a great resolution for skin cancer detection without the need for excision. One of these techniques, called photoacoustic imaging, uses the natural optical properties of some of our skin’s components, such as melanin, collagen, blood, or water. These skin’s elements emit a high-frequency acoustic signal when excited with light. This emission reveals the concentration of biological components in the sample analysed, just like a fingerprint. This can allow for the classification of healthy and tumorous tissue based on their characteristic composition.

However, due to the multiple tumour types and variations between patients, these techniques can be time-consuming and require human intervention for analysis and assessment. Introducing the use of artificial intelligence (AI) coupled with photoacoustic imaging can reduce the number of medical procedures and offer a less invasive and painful experience to the patient. Each AI model is trained on the singular tumour sample in order to eliminate the variance between different tumour types and
patient variations. These models can classify between healthy and tumorous pixels by inspecting their photoacoustic fingerprints and indicating the in-depth extension of the tumour. In this thesis, we demonstrate the efficacy of combining AI with photoacoustic imaging for skin cancer size prediction. Detecting skin tumours could become more accessible, meaning an earlier diagnosis, which is important for the effectiveness of the treatment and the prognosis of the disease. Such intelligent models could potentially offer unprecedented levels of accuracy in cancer extraction without the need for follow-up visits or the unnecessary removal of skin in case of wrongly diagnosed
tumours. Our technique can help the healthcare system to shift its focus to other crucial problems that still need human assessment and intervention. (Less)
Please use this url to cite or link to this publication:
author
Fracchia, Alice LU
supervisor
organization
course
FYTM05 20231
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Skin cancer, carcinoma, photoacoustic imaging, ultrasound imaging, machine learning, dimensionality reduction, sandpiles algorithm, active contour, multilayer perception (MLP), convolutional neural networks (CNN), autoencoder
language
English
id
9129566
date added to LUP
2023-06-27 12:00:50
date last changed
2023-08-31 03:42:15
@misc{9129566,
  abstract     = {{Skin cancer is a growing public health concern due to its prevalence among the population. Current clinical procedures require high invasiveness and multiple surgeries, which are responsible for patient discomfort and high medical expenses. Photoacoustic imaging offers an alternative to standard skin carcinoma diagnosis by exploiting the low scattering rate of ultrasound, which enables deep tissue penetration, and high imaging resolution. This thesis focuses on the application of machine learning models to systematically identify and delineate tumour regions according to their photoacoustic spectra. This is achieved through the implementation of a multilayer perceptron and a convolutional neural network that binary classify the spectral inputs. The same procedure is repeated on data that were dimensionally reduced by an autoencoder. The model predictions are further refined through post-processing contouring techniques. We apply our approach to six samples of the most common skin tumour types and successfully estimate the carcinoma extension. Specifically, the convolutional network accurately estimated the tumour extension, this way consistently
decreasing the size of excision margins. This model combined with the contouring technique constitutes a safe approach to skin cancer diagnosis. Our method significantly reduces the number of required surgeries and once automated will decrease the current medical staff workload.}},
  author       = {{Fracchia, Alice}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{A Machine Learning Approach to Skin Cancer Delineation on Photoacoustic Imaging}},
  year         = {{2023}},
}