Our colleagues from FORTH ( and Manos Koutoulakis) were in attendance at the 8th European Conference on Electrical Engineering and Computer Science 2024 in Bern, Switzerland. GenoMed4All was featured in two of their presentations for unsupervised bone marrow segmentation, radiocytogenetics of Multiple Myeloma (MM):
- A label-free ML method for segmenting the bone marrow in whole-body CT scans, using TotalSegmentator to detect the bone pixels, isolating the corresponding ROIs with bounding boxes, and applying unsupervised clustering to separate the bone from the bone marrow pixels.
- A novel radiocytogenetic model for matching the radiomic signature of bone marrow to the underlying chromosomal mutations. Thousands of imaging features were extracted from the segmented bone marrow and then linked to the cytogenetic score. The preliminary results are really promising and have the potential to address issues with the data completeness of multi-omic analyses.
Label-Free Machine Learning-based Segmentation of Whole-Body Bone Marrow Imaging in Multiple Myeloma
A clustering technique was adapted to identify the bone marrow pixels and morphological operations were employed to refine the segmentation mask. The qualitative analysis performed by experienced radiologists shows promising results. The proposed segmentation pipeline allows accurate and fast annotation of the whole-body bone marrow in multiple myeloma patients, achieving an IoU of 0.79±0.05 on the available cohort with femur bone annotations.
Radiocytogenetics in Multiple Myeloma: Predicting Cytogenetic Aberrations from WBCT Imaging Features
A machine learning analysis was employed to predict the expression of key chromosomal alterations and the cytogenetic risk of multiple myeloma patients. The proposed machine learning analysis based on sacrum and pelvis radiomics achieved the highest performance with an AUC of 0.76±0.03 among other radiocytogenetic models.