Table 1.

Summary of the presented concerns regarding AI application for glioblastoma MRI segmentation. Each limitation is accompanied by the domain of pertinence, the definition of the problem, and the proposed solution(s).

Section	Limitation	Domain	Definition	Possible Solution(s)
2.1	imaging heterogeneity	technical	scanner-dependent variation in image signal intensity	intensity standardization
				rescanning data
2.2	missing MRI sequences	technical	unavaiable modality/ies (T1, T2, FLAIR, T1CE)	inter-modality translation
				knowledge distillation
2.3	deployment issues	technical	limited computational resources and memory constraints	tiling
				quantization
2.4	performance evaluation	technical	subjective reference standards	cross-validation
				unsupervised training
3.1	limited number of patients	application	low number of data publicly avaiable	transfer learning
3.2	data quality	application	suboptimal quality of data (non-volumetric scans)	pre-processing
				inclusion of complex scenarios
3.3	data selection	application	selection bias and reduced applicability	inclusive database
3.4	focus on preoperative scenario	application	logistical and technical issues for postop. MRIs	multi-modality and multi-institutional data
4	exclusion of molecular data	molecular	limited consideration of IDH—1p/19q—MGMT	new coder architecture
				large-scale data-sharing
5.1	lack of standard guidelines	ethical	scientific integrity not definable	checklist
5.2	lack of transparency	ethical	limited understanding of the results	interpretability methods
				interdisciplinary collaboration
5.3	privacy and data protection	ethical	difficulty to obtain complete anonimization	skull-stripping