The hearing system is responsible for collecting, conducting, and amplifying sounds and converting them into electrical energy, and transmitting to specific centers in the brain [1]. The auditory organ consists of three parts: the outer ear, the middle ear, and the inner ear and the temporal bone is comprised of four parts including the squamous, mastoid, petrous, and tympanic parts. The petrous part surrounds the inner ear and the squamous part forms the mastoid appendage in the middle ear [2]. The mastoid appendage is formed by the attachment of the squamous part of the temporal bone [2]. Cells that form in different parts of the temporal bone all originate in the middle ear [3]. In general, the mastoid air cells are either pneumatized or none pneumatized. in the case of none pneumatization, it could have either opacification or sclerosis [4]. Mastoid air cells illustrate a comprehensive system of interlinking air-filled cavities surrounded by walls of the mastoid antrum and middle ear [5]. The mastoid part of the temporal bone has a significant role in terms of absorbance and scattering of kinetic energy through lateral trauma to the temporal bone, decreasing the occurrence of the fraction in the settling of direct trauma [6]. The concept of the grade of pneumatization of the temporal bone is so momentous in terms of surgical contemplations and pathophysiological care of numerous temporal bone illnesses [7]. Some range of the inflammatory, neoplastic, vascular, fibro-osseous, and traumatic changes have been illustrated by opacification at the middle ear and mastoid which help specialists to diagnose ear diseases [8]. One of the most prevalent complications of acute otitis media after tympanic membrane perforation is otomastoiditis which has risen over recent decades [9, 10]. Mastoiditis is an inflammation of the mastoid bone that is caused by inflammation of the middle ear and acute otitis due to the connection between the mastoid cells and the middle ear [11]. Because the middle ear is connected to the Eustachian tube on one side and to the mastoid cells by the aditus and anter on the other, whenever an infection reaches the middle ear and the tympanic membrane, this infection and inflammation may spread to the mastoid cells [12]. Therefore, the presence of mastoid pneumatic cells and the conjunction between the cells and the middle ear and the Eustachian tube is one of the proper ways that the infection and inflammation spread not only to the mastoid appendage but also to different parts of the temporal mastoid bone [12, 13]. In addition, the patients who were given chemotherapy or underwent organ transplantation surgery are mostly immunocompromised which caused the enhancement of the of otomastoiditis [9, 14]. If treatment for acute or chronic ear infection fails, the infection can spread to other areas of the head and neck. Even mastoid infections of the ear can be life-threatening disorders such as meningitis, subdural infections, brain abscesses, petrosal infections located between the inner and middle ears, temporal bone infections, and paralysis of the face [8]. The concept of the degree of pneumatization of the temporal bone is very important in terms of surgery and pathophysiological care of many temporal bone diseases [8]. A wide range of inflammatory, neoplastic, vascular, fibrous, and traumatic changes with opacification in the middle ear and mastoid have been shown to assist specialists in diagnosing ear disease [8]. Sclerosis and opacification of the middle ear and mastoid air cells are key CT features of various ear diseases including acute otomastoiditis, necrotizing otitis externa, chronic otomastoiditis and cholesteatoma [8]. There are some other key CT features of mentioned ear diseases, for instance: CT features of acute otomastoiditis are middle ear and mastoid opacification with liquid levels and probably bone demolition [8]. Vast soft-tissue inflammation with middle ear and mastoid opacification and skull base osteomyelitis leading to bony demolition is the key CT features of necrotizing otitis externa [8]. For chronic otomastoiditis there are middle ear and mastoid opacification with mastoid trabeculae inspissating, sclerosis, and cell sabotage, probably ossicular chain abrasion [8]. Also, cholesteatoma causes middle ear and/or mastoid soft-tissue opacification with ossicular, caul tympani, or scutum abrasion, probably labyrinthine fistulas [8]. Mastoid process involved in some other diseases for instance, covid19 [15]. There is a poor correlation between mastoiditis on CT imaging with the clinical diagnosis which emphasizes the importance of CT images [13]. Clinical intervention for opacification in the mastoid process is very crucial after a diagnosis [16]. In some cases, we witness the discrepancies between CT reports and surgical findings regarding middle ear opacification which is mostly caused by misdetection of radiologists in imaging [17].

To find the solution for usual issues in clinical actions such as eruptively radiologists’ workloads and innate challenges of explicating of medical images, the usage of deep learning has been explored by many studies in order to find the best model in terms of analyzing them.

There are so many achievements by using deep learning methods in diverse functions of computer vision for instance image classifications, object recognition, localization, and segmentation in natural images [18]. Convolutional Neural Networks (CNN) are used in medical images for detecting and evaluating illnesses [18]. Diagnosis of unique traits of medical images is customarily carried out by experts for detecting diseases. Neural networks or deep learning in various medical fields have shown great success. Automated diagnosis by artificial intelligence has recently been the focus of specialist physicians due to the significant decrement in error and high speed of diagnosis [18]. In some fields, the results were excellent compared to those of specialists [19, 20]. The main problem with generating CNN is its need for large number of training data which is not feasible in most cases [21]. Alternatively, pre-trained public CNN models for natural images could be utilized and fine-tuned to a particular usage which is named Transfer learning [21]. Transfer learning is the concept of dominating the cloistered learning template and using knowledge obtained for one task and solving related ones. In transfer learning, most of the network layers are transferred to the new model. But the difference is in the Fully-connected layers which are changed based on the new set of classes [21]. Previous studies have shown that the use of transfer learning in medical imaging has better results than building CNN from scratch [19, 21].

Some recent studies were performed for automatic diagnosis of ear diseases by using endoscopic images. A study was conducted in for diagnosing otitis media and they got 81.58% accuracy via decision and 86.84% via neural networks [22]. The other study performed by Cha et al. [23] in which the otoendoscopic images were used via public convolution-based deep neural networks to categorize common ear illnesses (Normal, Attic retraction, Tympanic perforation, Otitis externa ± myringitis, Tumor) [23]. Some ear disorders only can be diagnosed by computed tomography scans. The most relevant study conducted regarding mastoid abnormalities was introduced in radiographic images [24]. Since the mastoiditis incidence is mostly occurred in under two years old children who are very susceptible to radiation exposure [24]. With using multiple views, the area under the curve of their proposed algorithm was 0.971, 0.978, and 0.965 for the gold standard, temporal, and geographic external test sets, respectively [24]. And also the sensitivity and specificity of their method were 96.4% and 74.5% respectively [24]. However, the most detailed abnormalities in tiny parts of the middle ear such as petrosal and sigmoid sinus can’t be detected in the radiographic images, and the use of radiography has obsoleted [25]. The most general technique utilized to elicit the details of images of the ear cavity is computed tomography scan (CT scan) [25]. The processing of mastoid air cells is only partially represented on a CT scan by Olivier Cros [26]. In [27] a two-class (normal and abnormal) classifier based on convolutional neural networks deep learning model was introduced. The proposed model has an accuracy of 98.10%, however, this study classifies only normal and abnormal mastoids.

In this paper the first and robust deep learning-based approaches is introduced to diagnose mastoid abnormalities in five groups (1. Complete pneumatization, 2. Opacification in pneumatization, 3. Partial pneumatization, 4. Opacification in partial pneumatization, 5. None pneumatized). The proposed method can reduce the analysis of the large and complex CT images which may be a tedious and complex task for clinicians.

This paper is organized as follows. In Section “Materials and methods”, we explain the used dataset and also our proposed deep learning-based method. The results and performance evaluation are presented in the “Results” section. Finally, the paper is concluded in the “Conclusion and discussion” Sections.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Disclaimer:

This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (https://www.springeropen.com/)