Comparison of semi-automatic and manual segmentation methods for tumour delineation on Head and Neck Squamous Cell Carcinoma (HNSCC) PET images.

Mahima Merin Philip, Andy Welch, Fergus McKiddie, Mintu Nath

Research output: Contribution to conferencePosterpeer-review


An accurate and reproducible tumour delineation on Positron Emission Tomography (PET) images is required to validate predictive and prognostic models based on PET features. The manual segmentation of tumour on PET images by an expert is the acceptable standard, but it is time-consuming. On the other hand, semi-automatic methods are easily implementable, rapid and inexpensive. This study assessed the reliability of semi-automatic segmentation methods over manual segmentation for tumour delineation in Head and Neck Squamous Cell Carcinoma (HNSCC) PET images. We employed manual and four semi-automatic segmentation methods (LOGISMOS based approach, watershed, grow from seeds (GfS), 40%SUVmax threshold) to extract 128 radiomic features from FDG-PET images of 70 patients. The semi-automatic methods were implemented using 3D slicer software. All features were imputed by their median values, evaluated for their distributional properties, and log-transformed when they showed the right skewness. We fitted a separate linear mixed effect model for each feature using the method as a fixed effect and patient as a random effect. We estimated the intraclass correlation coefficient (ICC) from the fitted model and summarised the outcomes to reflect the consistency and agreement between the manual and semi-automatic methods. We also explored other statistics like limits of agreement, total deviation index and coverage probability to evaluate the agreement between methods. Most features obtained using watershed and GfS showed higher consistency (0.36 to 0.99) and agreement (0.35 to 0.99) with manually segmented tumours. Features based on the grey level co-occurrence matrix (GLCM) and neighbouring grey tone difference matrix (NGTDM) for the watershed method and GLCM for the Gfs method showed the best estimates of consistency with the manual method (all 0.99). Other statistics also presented similar outcomes. We conclude that both watershed and GfS are reliable semi-automatic methods for tumour delineation on HNSCC PET images.
Original languageEnglish
Publication statusE-pub ahead of print - 12 Sep 2022
EventRSS International Conference 2022 - P&J Live, Aberdeen, United Kingdom
Duration: 12 Sep 202215 Sep 2022


ConferenceRSS International Conference 2022
Country/TerritoryUnited Kingdom
Internet address


  • Tumour
  • Image
  • Segmentation methods
  • Interclass correlation
  • agreement


Dive into the research topics of 'Comparison of semi-automatic and manual segmentation methods for tumour delineation on Head and Neck Squamous Cell Carcinoma (HNSCC) PET images.'. Together they form a unique fingerprint.

Cite this