Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Advancing Precision Diagnostics for Delineation of Basal Cell Carcinoma in Hyperspectral Images

Ibek, Paulina LU (2025) In Master’s Theses in Mathematical Sciences BERM03 20242
Centre for Mathematical Sciences
Abstract
Basal cell carcinoma (BCC) is the most common form of skin cancer with the current diagnostic procedures relying on surgical excisions and histopathological analyses. Incomplete removal of the tumor necessitates additional surgery, which is both costly and time-consuming, and causes unnecessary suffering for the patient. To address this, non-invasive diagnostic techniques capable of determining the dimensions of skin cancer preoperatively are urgently needed. Furthermore, there also exists no automatic framework to do this, and all suggested methods require histopathological input with manual delineation.
This thesis presents the development of an automated diagnostic pipeline for delineating basal cell carcinoma, using hyperspectral... (More)
Basal cell carcinoma (BCC) is the most common form of skin cancer with the current diagnostic procedures relying on surgical excisions and histopathological analyses. Incomplete removal of the tumor necessitates additional surgery, which is both costly and time-consuming, and causes unnecessary suffering for the patient. To address this, non-invasive diagnostic techniques capable of determining the dimensions of skin cancer preoperatively are urgently needed. Furthermore, there also exists no automatic framework to do this, and all suggested methods require histopathological input with manual delineation.
This thesis presents the development of an automated diagnostic pipeline for delineating basal cell carcinoma, using hyperspectral imaging (HSI), combined with computational methods. HSI is a non-invasive, high-resolution imaging technique capable of differentiating different tissue structures on the basis of their molecular composition. The proposed pipeline involves extensive data preprocessing, followed by machine learning to identify tumors in hyperspectral data. It is designed to be retrained and adapted for individual patients, as no universal characteristic signal exists for skin cancer, a finding supported in this study. The pipeline was evaluated on a sample population of 21 individuals, where the predictions were compared to histopathology achieving a strong correlation (average R^2= 0.86). By comparing spectral signatures between healthy and diseased tissue at an individual level, the method enables personalized diagnostics and precise delineation of tumor boundaries. Such identification would not be possible to achieve with group-level analysis, where individual variability obscures tumor-specific signals. The thesis also suggests that white reference normalization—a cumbersome step in hyperspectral setups—can be bypassed, with normalization handled during processing, thereby simplifying the measurement process. These findings demonstrate promising results for facilitating more efficient procedures for skin tumor diagnostics, with direct implications for improving patient care. (Less)
Popular Abstract (Swedish)
Basalcellscancer (BCC) är den vanligaste formen av hudcancer som enligt Socialstyrelsen drabbar omkring 50 000 svenskar varje år. Om en patient misstänks ha basalcellscancer måste man ta bort en bit av huden för att undersöka den i mikroskop. Om det visar sig efter undersökningen att det finns cancerceller kvar, måste man operera igen, vilket är obehagligt för patienten, kostsamt och förlänger väntetider. Därför vore det bra om det fanns ett sätt att mäta hur stor tumören är innan operationen. Dessutom vore det bra om det fanns ett automatiskt system för att hjälpa till med diagnostiken – exempelvis en dator med ett program som kan visa hur stor tumören är. Inga sådana metoder finns idag, utan all diagnostik kräver manuell bedömning av en... (More)
Basalcellscancer (BCC) är den vanligaste formen av hudcancer som enligt Socialstyrelsen drabbar omkring 50 000 svenskar varje år. Om en patient misstänks ha basalcellscancer måste man ta bort en bit av huden för att undersöka den i mikroskop. Om det visar sig efter undersökningen att det finns cancerceller kvar, måste man operera igen, vilket är obehagligt för patienten, kostsamt och förlänger väntetider. Därför vore det bra om det fanns ett sätt att mäta hur stor tumören är innan operationen. Dessutom vore det bra om det fanns ett automatiskt system för att hjälpa till med diagnostiken – exempelvis en dator med ett program som kan visa hur stor tumören är. Inga sådana metoder finns idag, utan all diagnostik kräver manuell bedömning av en patolog. I detta examensarbete presenteras en lösning där en automatiserad metod och analys utvecklats för att avgränsa basalcellscancer i hyperspektala bilder (HSI). HSI är en avancerad teknik som kan skapa bilder av huden med information som inte kan uppfattas av våra ögon, vilket gör det möjligt att skilja på olika vävnadstyper. HSI åstadkommer detta genom att analysera vävnadens molekylära sammansättning via så kallade spektrala signaler. HSI kan alltså ta en bild som kan användas för att skilja på cancer och vanlig hud, vilket man inte kan göra genom att titta med bara ögat. I examensarbetet används olika statistiska metoder och maskininlärning för att identifiera tumörer baserat på hyperspektral data. Det som gör algoritmen unik är att den anpassas för varje individuell patient, eftersom det inte finns ett universellt mönster för hudcancer. Dessutom är algoritmen självgående, och kan hitta tumörens gränser utan läkarhjälp. Metoden testades på 21 personer, och resultaten visade en stark överensstämmelse med de traditionella analyserna som görs i laboratorier idag. Om algoritmen utvecklas ytterligare så kan den i framtiden användas på kliniker för att hjälpa läkare bedöma hur mycket hud de måste skära bort. Tumören måste i så fall fortfarande bedömas histopatologiskt, men förhoppningen är att antalet operationer går ner, och väntetiden blir kortare. (Less)
Please use this url to cite or link to this publication:
author
Ibek, Paulina LU
supervisor
organization
course
BERM03 20242
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Basal cell carcinoma, machine learning, precision diagnostics, unsupervised learning, PCA, K-means clustering, hyperspectral imaging, AI, cancer, image analysis
publication/series
Master’s Theses in Mathematical Sciences
report number
LUNFBV-3001-2025
ISSN
1404-6342
other publication id
2025:E12
language
English
id
9184068
date added to LUP
2025-10-22 15:46:02
date last changed
2025-10-22 15:46:02
@misc{9184068,
  abstract     = {{Basal cell carcinoma (BCC) is the most common form of skin cancer with the current diagnostic procedures relying on surgical excisions and histopathological analyses. Incomplete removal of the tumor necessitates additional surgery, which is both costly and time-consuming, and causes unnecessary suffering for the patient. To address this, non-invasive diagnostic techniques capable of determining the dimensions of skin cancer preoperatively are urgently needed. Furthermore, there also exists no automatic framework to do this, and all suggested methods require histopathological input with manual delineation. 
This thesis presents the development of an automated diagnostic pipeline for delineating basal cell carcinoma, using hyperspectral imaging (HSI), combined with computational methods. HSI is a non-invasive, high-resolution imaging technique capable of differentiating different tissue structures on the basis of their molecular composition. The proposed pipeline involves extensive data preprocessing, followed by machine learning to identify tumors in hyperspectral data. It is designed to be retrained and adapted for individual patients, as no universal characteristic signal exists for skin cancer, a finding supported in this study. The pipeline was evaluated on a sample population of 21 individuals, where the predictions were compared to histopathology achieving a strong correlation (average R^2= 0.86). By comparing spectral signatures between healthy and diseased tissue at an individual level, the method enables personalized diagnostics and precise delineation of tumor boundaries. Such identification would not be possible to achieve with group-level analysis, where individual variability obscures tumor-specific signals. The thesis also suggests that white reference normalization—a cumbersome step in hyperspectral setups—can be bypassed, with normalization handled during processing, thereby simplifying the measurement process. These findings demonstrate promising results for facilitating more efficient procedures for skin tumor diagnostics, with direct implications for improving patient care.}},
  author       = {{Ibek, Paulina}},
  issn         = {{1404-6342}},
  language     = {{eng}},
  note         = {{Student Paper}},
  series       = {{Master’s Theses in Mathematical Sciences}},
  title        = {{Advancing Precision Diagnostics for Delineation of Basal Cell Carcinoma in Hyperspectral Images}},
  year         = {{2025}},
}