Abstract
Classification of biological images plays a crucial role in many biological problems, e.g. recognition of cell phenotypes and maturation levels, localization of cell organelles and histopathological classification, and holds the potential to support early diagnosis, which is critical in disease prevention. In this paper, we tested different ensemble of canonical and deep classifiers to provide accurate identification of actinic keratosis (AK), one of the most common skin lesions that could degenerate into lethal squamous cell carcinomas. We used a clinical image dataset to build and test different ensembles of support vector machines trained by handcrafted descriptors and convolutional neural networks (CNNs) for which we experimented different learning rates, augmentation techniques (e.g. warping) and topologies. Our results show that the proposed ensemble obtains performance comparable to the state of the art. To reproduce the experiments reported in this paper, the MATLAB code of all the descriptors is available at https://github.com/LorisNanni.
Highlights
Automated classification of biological images by means of computer vision can be applied with great success to a variety of biological problems – examples include organelle classification and location, assessment of maturation level of cells, and tissue and cancer recognition [1]
To reproduce the method proposed in this paper, the MATLAB code is available at https://github.com/LorisNanni
We used the dataset shared by Spyridonos et al [9,10], that contains 6010 control and 16269 actinic keratosis (AK) 50 ×50 pixels regions of interest (ROIs) from clinical photographs acquired from 22 patients
Summary
Automated classification of biological images by means of computer vision can be applied with great success to a variety of biological problems – examples include organelle classification and location, assessment of maturation level of cells, and tissue and cancer recognition [1]. The gold standard for diagnosing SCC is skin biopsy, but several non-invasive alternatives are common in the clinical practice: optical coherence tomography, dermoscopy, reflectance confocal microscopy or stripping mRNA are less intrusive for the patient. The drawback for such techniques is that training and experience of the clinicians play a key role in generating the correct diagnosis [5]. The proof that an ensemble of different CNNs trained with different data augmentation strategies outperforms the single networks and the handcrafted baselines. To reproduce the method proposed in this paper, the MATLAB code is available at https://github.com/LorisNanni
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call