A retrospective study of deep learning generalization across two centers and multiple models of X-ray devices using COVID-19 chest-X rays.

Generalization of deep learning (DL) algorithms is critical for the secure implementation of computer-aided diagnosis systems in clinical practice. However, broad generalization remains to be a challenge in machine learning. This research aims to identify and study potential factors that can affect the internal validation and generalization of DL networks, namely the institution where the images come from, the image processing applied by the X-ray device, and the type of response function of the X-ray device. For these purposes, a pre-trained convolutional neural network (CNN) (VGG16) was trained three times for classifying COVID-19 and control chest radiographs with the same hyperparameters, but using different combinations of data acquired in two institutions by three different X-ray device manufacturers. Regarding internal validation, the addition of images from an external institution to the training set did not modify the algorithm's internal performance, however, the inclusion of images acquired by a device from a different manufacturer decreased the performance up to 8% (p < 0.05). In contrast, generalization across institutions and X-ray devices with the same type of response function was achieved. Nonetheless, generalization was not observed across devices with different types of response function. This factor was the key impediment to achieving broad generalization in our research, followed by the device's image-processing and the inter-institutional differences, which both reduced generalization performance to 18.9% (p < 0.05), and 9.8% (p < 0.05), respectively. Finally, clustering analysis with features extracted by the CNN was performed, revealing a substantial dependence of feature values extracted by the pre-trained CNN on the X-ray device which acquired the images.

Histological Reaction in the Round Window Membrane after Cochlear Implant Insertion in Nonhuman Primates.

INTRODUCTION: The primary objective of this article was to determine if any histological alterations occur in the round window (RW) and adjacent anatomical structures of an animal model with normal hearing when a cochlear implant (CI) electrode array is inserted. Furthermore, this article aims to relate these histological alterations to the auditory changes generated. METHODS: Cochlear implantation was performed, following the principles of minimally traumatic surgery (MTS), in 15 ears of nonhuman primates (Macaca fascicularis) (Mf) with normal hearing. Auditory brainstem-evoked potentials (ABR) using clicks and tones were used prior to surgery and during a 6-month follow-up period. Histological evaluation was carried out, analyzing the position of the electrode array with respect to the round window membrane (RWM), its percentage of occupation and integrity, the presence of intracochlear damage, and the tissue reaction provoked, the latter of which was quantified in the temporal bones. RESULTS: Surgery was performed on all 15 ears without relevant incidences. Regarding histology, the electrode array in the RW of all ears presented a lateral position with respect to the modiolus. No lesions affecting the integrity of the RW were observed. The mean value of the array's occupation in the RW was 25%. Tissue reaction, in the form of fibrosis, was observed in all ears and more intensely in the trans-RWM and post-RWM areas. In all ears, the electrodes remained on the scala tympani. No profound hearing impairment was recorded in any ear, being the mean loss of 25.4 dB when comparing presurgical thresholds with those collected 6 months after implantation in ABR click and 24.4 dB in ABR tone burst. CONCLUSIONS: The animal model and Hybrid L-14 (HL14) electrode array were optimal for implementing a surgical technique similar to that routinely performed on humans. Mild histological alterations were observed in the round window membrane and adjacent anatomical structures from the insertion of a cochlear implant electrode array. Following the minimally invasive technique, levels of hearing preservation were satisfactory, reaching a pre-post difference of 25.4 dB in the ABR click and 24.4 dB for a high-frequency tone burst. Complete hearing impairment was not observed in either ear. Correlation between the severity of histological alterations and hearing changes recorded in the ABR studies was observed.