Table 5. Overview of classification approaches for fibred endoscopic imaging going beyond traditional machine learning.

Organ (System)	Classifying	References	Methodology	Comments
Pulmonary	Cancerous nodules in airways.	Gil et al. (2017)	Unsupervised classification (compensating for limited data availability) using graph representation and community detection algorithms.	Early FBEμ classification approaches going beyond the traditional machine learning pipeline, exploring methods such as Convolutional Neural Networks (off the self as well as custom), transfer learning, unsupervised learning and multi-modal learning at a latent space. The results are very promising. Yet, more data, both in terms of numbers as well as in terms of diversity are necessary. Furthermore, custom solutions, taking into consideration the inherent FBEμ imaging properties, could further enhance the classification performance.
Gastro-intestinal	Oesophagus epithelial changes.	Hong et al. (2017) and Aubreville et al. (2017)	Custom CNN architecture for the multi-class frame classification.
Oropharyngeal	Pathological epithelium.	Aubreville et al. (2017)	Full-training of LeNet-5 and shallow fine-tuning the Inception v3 (using the ImageNet database).
Brain	Informative frames within videos.	Izadyyazdanabadi et al. (2017), Izadyyazdanabadi et al. (2018)	Fully-trained AlexNet and GoogleNet as well as comparing the between full training and transfer learning through fine-tuning using the ImageNet database.
	Brain tumours.	Murthy et al. (2017)	Novel Cascaded CNN, discarding easy images at early stages, concentrating on challenging ones at subsequent, expert shallow nets.
Breast	Cancerous breast nodules.	Gu et al. (2017)	Multi-modal (FBEμ mosaics and histology) classification mapping the original features to a latent space for improved SVM performance.