Image diagnosis models for the oral assessment of older people using convolutional neural networks: A retrospective observational study

AIMS

The purpose of this study was to construct a model for oral assessment using deep learning image recognition technology and to verify its accuracy.

BACKGROUND

The effects of oral care on older people are significant, and the Oral Assessment Guide has been used internationally as an effective oral assessment tool in clinical practice. However, additional training, education, development of user manuals and continuous support from a dental hygienist are needed to improve the inter-rater reliability of the Oral Assessment Guide.

DESIGN

A retrospective observational study.

METHODS

A total of 3,201 oral images of 114 older people aged >65 years were collected from five dental-related facilities. These images were divided into six categories (lips, tongue, saliva, mucosa, gingiva, and teeth or dentures) that were evaluated by images, out of the total eight items that comprise components of the Oral Assessment Guide. Each item was classified into a rating of 1, 2 or 3. A convolutional neural network, which is a deep learning method used for image recognition, was used to construct the image recognition model. The study methods comply with the STROBE checklist.

RESULTS

We constructed models with a classification accuracy of 98.8% for lips, 94.3% for tongue, 92.8% for saliva, 78.6% for mucous membranes, 93.0% for gingiva and 93.6% for teeth or dentures.

CONCLUSIONS

Highly accurate diagnostic imaging models using convolutional neural networks were constructed for six items of the Oral Assessment Guide and validated. In particular, for the five items of lips, tongue, saliva, gingiva, and teeth or dentures, models with a high accuracy of over 90% were obtained.

RELEVANCE TO CLINICAL PRACTICE

The model built in this study has the potential to contribute to obtain reproducibility and reliability of the ratings, to shorten the time for assessment, to collaborate with dental professionals and to be used as an educational tool.

Deep Learning Approach in Image Diagnosis of Pseudomonas Keratitis

This investigation aimed to explore deep learning (DL) models' potential for diagnosing Pseudomonas keratitis using external eye images. In the retrospective research, the images of bacterial keratitis (BK, n = 929), classified as Pseudomonas (n = 618) and non-Pseudomonas (n = 311) keratitis, were collected. Eight DL algorithms, including ResNet50, DenseNet121, ResNeXt50, SE-ResNet50, and EfficientNets B0 to B3, were adopted as backbone models to train and obtain the best ensemble 2-, 3-, 4-, and 5-DL models. Five-fold cross-validation was used to determine the ability of single and ensemble models to diagnose Pseudomonas keratitis. The EfficientNet B2 model had the highest accuracy (71.2%) of the eight single-DL models, while the best ensemble 4-DL model showed the highest accuracy (72.1%) among the ensemble models. However, no statistical difference was shown in the area under the receiver operating characteristic curve and diagnostic accuracy among these single-DL models and among the four best ensemble models. As a proof of concept, the DL approach, via external eye photos, could assist in identifying Pseudomonas keratitis from BK patients. All the best ensemble models can enhance the performance of constituent DL models in diagnosing Pseudomonas keratitis, but the enhancement effect appears to be limited.

Case report: Misdiagnosis of accessory spleen in the left adrenal region as an adrenal tumour after splenectomy

Background

Adrenal tumours are common in urology and endocrinology, and the diagnosis of adrenal tumours were mainly depends on imaging diagnosis. Howerver, misdiagnosis can still occur for some adrenal space-occupying lesions without specific manifestations or abnormal biochemical indexes.

Methods

We report the case of a 55-year-old patient with a soft-tissue mass in the left adrenal region, and have no specific manifestations or abnormalities in biochemical indexes. The patient had undergone open splenectomy 20 years ago for splenic rupture caused by traffic-accident trauma, and had a 10-year special history of hypertension. Because of the uncertain nature of the mass, surgical treatment was recommended.

Results

The surgeon managed to remove the left adrenal region mass. During the surgery, the adrenal source was excluded. In the histological examination, the splenic corpuscle and splenic medullary structure were seen under the microscope, and an accessory spleen was diagnosed.

Conclusions

The accessory spleen was located in the adrenal region rarely, and can easily be misdiagnosed as an adrenal tumour. When the cases show abnormal adrenal space-occupying lesions in imaging examinations, non-adrenal diseases should be considered. we need to combine different imaging techniques for analysis, and think more about it, avoid misdiagnosis leading to unnecessary surgery.

Deep Convolutional Neural Network for Nasopharyngeal Carcinoma Discrimination on MRI by Comparison of Hierarchical and Simple Layered Convolutional Neural Networks

Nasopharyngeal carcinoma (NPC) is one of the most common head and neck cancers. Early diagnosis plays a critical role in the treatment of NPC. To aid diagnosis, deep learning methods can provide interpretable clues for identifying NPC from magnetic resonance images (MRI). To identify the optimal models, we compared the discrimination performance of hierarchical and simple layered convolutional neural networks (CNN). Retrospectively, we collected the MRI images of patients and manually built the tailored NPC image dataset. We examined the performance of the representative CNN models including shallow CNN, ResNet50, ResNet101, and EfficientNet-B7. By fine-tuning, shallow CNN, ResNet50, ResNet101, and EfficientNet-B7 achieved the precision of 72.2%, 94.4%, 92.6%, and 88.4%, displaying the superiority of deep hierarchical neural networks. Among the examined models, ResNet50 with pre-trained weights demonstrated the best classification performance over other types of CNN with accuracy, precision, and an F1-score of 0.93, 0.94, and 0.93, respectively. The fine-tuned ResNet50 achieved the highest prediction performance and can be used as a potential tool for aiding the diagnosis of NPC tumors.

An artificial intelligence-assisted diagnostic system improves the accuracy of image diagnosis of uterine cervical lesions

The present study created an artificial intelligence (AI)-automated diagnostics system for uterine cervical lesions and assessed the performance of these images for AI diagnostic imaging of pathological cervical lesions. A total of 463 colposcopic images were analyzed. The traditional colposcopy diagnoses were compared to those obtained by AI image diagnosis. Next, 100 images were presented to a panel of 32 gynecologists who independently examined each image in a blinded fashion and diagnosed them for four categories of tumors. Then, the 32 gynecologists revisited their diagnosis for each image after being informed of the AI diagnosis. The present study assessed any changes in physician diagnosis and the accuracy of AI-image-assisted diagnosis (AISD). The accuracy of AI was 57.8% for normal, 35.4% for cervical intraepithelial neoplasia (CIN)1, 40.5% for CIN2-3 and 44.2% for invasive cancer. The accuracy of gynecologist diagnoses from cervical pathological images, before knowing the AI image diagnosis, was 54.4% for CIN2-3 and 38.9% for invasive cancer. After learning of the AISD, their accuracy improved to 58.0% for CIN2-3 and 48.5% for invasive cancer. AI-assisted image diagnosis was able to improve gynecologist diagnosis accuracy significantly (P<0.01) for invasive cancer and tended to improve their accuracy for CIN2-3 (P=0.14).

Narrative review of generative adversarial networks in medical and molecular imaging

Recent years have witnessed a rapidly expanding use of artificial intelligence and machine learning in medical imaging. Generative adversarial networks (GANs) are techniques to synthesize images based on artificial neural networks and deep learning. In addition to the flexibility and versatility inherent in deep learning on which the GANs are based, the potential problem-solving ability of the GANs has attracted attention and is being vigorously studied in the medical and molecular imaging fields. Here this narrative review provides a comprehensive overview for GANs and discuss their usefulness in medical and molecular imaging on the following topics: (I) data augmentation to increase training data for AI-based computer-aided diagnosis as a solution for the data-hungry nature of such training sets; (II) modality conversion to complement the shortcomings of a single modality that reflects certain physical measurement principles, such as from magnetic resonance (MR) to computed tomography (CT) images or vice versa; (III) de-noising to realize less injection and/or radiation dose for nuclear medicine and CT; (IV) image reconstruction for shortening MR acquisition time while maintaining high image quality; (V) super-resolution to produce a high-resolution image from low-resolution one; (VI) domain adaptation which utilizes knowledge such as supervised labels and annotations from a source domain to the target domain with no or insufficient knowledge; and (VII) image generation with disease severity and radiogenomics. GANs are promising tools for medical and molecular imaging. The progress of model architectures and their applications should continue to be noteworthy.

Image-based diagnosis of residual or recurrent nasopharyngeal carcinoma may be a phantom tumor phenomenon

Some nasopharyngeal carcinoma (NPC) patients may present convincing radiological evidence mimicking residual or recurrent tumor after radiotherapy. However, by means of biopsies and long term follow-up, the radiologically diagnosed residuals/recurrences are not always what they appear to be. We report our experience on this "phantom tumor" phenomenon. This may help to avoid the unnecessary and devastating re-irradiation subsequent to the incorrect diagnosis.In this longitudinal cohort study, we collected 19 patients of image-based diagnosis of residual/recurrent NPC during the period from Feb, 2010 to Nov. 2016, and then observed them until June, 2019. They were subsequently confirmed to have no residual/recurrent lesions by histological or clinical measures. Image findings and pathological features were analyzed.Six patients showed residual tumors after completion of radiotherapy and 13 were radiologically diagnosed to have recurrences based on magnetic resonance imaging (MRI) criteria 6 to 206 months after radiotherapy. There were 3 types of image patterns: extensive recurrent skull base lesions (10/19); a persistent or residual primary lesion (3/19); lesions both in the nasopharynx and skull base (6/19). Fourteen patients had biopsy of the lesions. The histological diagnoses included necrosis/ inflammation in 10 (52.7%), granulation tissue with inflammation in 2, and reactive epithelial cell in 1. Five patients had no pathological proof and were judged to have no real recurrence/residual tumor based on the absence of detectable plasma EB virus DNA and subjective judgment. These 5 patients have remained well after an interval of 38-121 months without anti-cancer treatments.Image-based diagnosis of residual or recurrent nasopharyngeal carcinoma may be unreliable. False positivity, the "phantom tumor phenomenon", is not uncommon in post-radiotherapy MRI. This is particularly true if the images show extensive skull base involvement at 5 years or more after completion of radiotherapy. MRI findings compatible with NPC features must be treated as a real threat until proved otherwise. However, the balance between under- and over-diagnosis must be carefully sought. Without a pathological confirmation, the diagnosis of residual or recurrent NPC must be made taking into account physical examination results, endoscopic findings and Epstein-Barr virus viral load. A subjective medical judgment is needed based on clinical and laboratory data and the unique anatomic complexities of the nasopharynx.

Development of a light-weight deep learning model for cloud applications and remote diagnosis of skin cancers

Skin cancer is among the 10 most common cancers. Recent research revealed the superiority of artificial intelligence (AI) over dermatologists to diagnose skin cancer from predesignated and cropped images. However, there remain several uncertainties for AI in diagnosing skin cancers, including lack of testing for consistency, lack of pathological proof or ambiguous comparisons. Hence, to develop a reliable, feasible and user-friendly platform to facilitate the automatic diagnostic algorithm is important. The aim of this study was to build a light-weight skin cancer classification model based on deep learning methods for aiding first-line medical care. The developed model can be deployed on cloud platforms as well as mobile devices for remote diagnostic applications. We reviewed the medical records and clinical images of patients who received a histological diagnosis of basal cell carcinoma, squamous cell carcinoma, melanoma, seborrheic keratosis and melanocytic nevus in 2006-2017 in the Department of Dermatology in Kaohsiung Chang Gung Memorial Hospital (KCGMH). We used the deep learning models to identify skin cancers and benign skin tumors in the manner of binary classification and multi-class classification in the KCGMH and HAM10000 datasets to construct a skin cancer classification model. The accuracy reached 89.5% for the binary classifications (benign vs malignant) in the KCGMH dataset; the accuracy was 85.8% in the HAM10000 dataset in seven-class classification and 72.1% in the KCGMH dataset in five-class classification. Our results demonstrate that our skin cancer classification model based on deep learning methods is a highly promising aid for the clinical diagnosis and early identification of skin cancers and benign tumors.

Dermoscopic diagnostic performance of Japanese dermatologists for skin tumors differs by patient origin: A deep learning convolutional neural network closes the gap

In the dermoscopic diagnosis of skin tumors, it remains unclear whether a deep neural network (DNN) trained with images from fair-skinned-predominant archives is helpful when applied for patients with darker skin. This study compared the performance of 30 Japanese dermatologists with that of a DNN for the dermoscopic diagnosis of International Skin Imaging Collaboration (ISIC) and Shinshu (Japanese only) datasets to classify malignant melanoma, melanocytic nevus, basal cell carcinoma and benign keratosis on the non-volar skin. The DNN was trained using 12 254 images from the ISIC set and 594 images from the Shinshu set. The sensitivity for malignancy prediction by the dermatologists was significantly higher for the Shinshu set than for the ISIC set (0.853 [95% confidence interval, 0.820-0.885] vs 0.608 [0.553-0.664], P < 0.001). The specificity of the DNN at the dermatologists' mean sensitivity value was 0.962 for the Shinshu set and 1.00 for the ISIC set and significantly higher than that for the human readers (both P < 0.001). The dermoscopic diagnostic performance of dermatologists for skin tumors tended to be less accurate for patients of non-local populations, particularly in relation to the dominant skin type. A DNN may help close this gap in the clinical setting.

Dermoscopy image-based self-learning on computer improves diagnostic performance of medical students compared with classroom-style lecture in ultra-short period

The educational effectiveness of dermoscopy image-based self-learning on a computer for medical students has not been well examined. To assess the effect of an image-based self-learning session on the dermoscopic diagnostic performance for malignant melanoma (MM), basal cell carcinoma, melanocytic nevus and seborrheic keratosis (SK) on non-acral regions in comparison with a conventional classroom-style lecture, 114 fourth-year medical students (mean age, 23.7 years; male : female, 73:41) were enrolled. The subjects were randomly assigned to either a self-learning to lecture (SL) or lecture to self-learning (LS) group to receive a 15-min image-based self-learning computer session and a 15-min video lecture session in different orders. The user interface of the digital content was the same as that on a website (https://dz-image.casio.jp). Diagnostic performance was determined using the total number of correct answers for the four diseases and by malignancy prediction in examination A (before training), B (after receiving one session) and C (after receiving both sessions). The examinations were all unique and contained five dermoscopic images each of the four diseases. The total number of correct answers and malignancy prediction results for examination B were significantly higher in the SL group than in LS (11.6 and 15.2 vs 10.1 and 13.4, respectively; both P < 0.01), with no remarkable differences for examination C (13.5 and 16.8 vs 13.3 and 16.4, respectively; P = 0.62 and P = 0.21). In subanalyses, the number of correct answers for SK in examination B was significantly higher in the SL group (3.6 vs. 1.8, P < 0.01), while that for MM was significantly lower (2.2 vs 3.0, P < 0.01). Diagnostic performance was comparable between sexes for examination B. In conclusion, computer-assisted dermoscopy image-based self-learning may be a suitable and non-inferior alternative to classroom-style instruction for medical students within an ultra-short training period.

[Do we agree on the technical quality of our chest X-rays? A radiologist, an expert in X-ray diagnosis, and a family doctor try to find out]

OBJECTIVE

To appraise the degree of concordance in the interpretation of the technical quality of chest x-rays at a health centre between an expert in x-ray diagnosis, a family doctor, and a radiologist.

DESIGN

Transversal study. Setting. Primary care. Cartuja Health Centre, Granada, Spain.

PARTICIPANTS

Patients at the Cartuja Health Centre who had a simple chest x-ray in 2002. 150 studies were chosen by simple randomised sampling. Two were rejected because they dealt with a repeat examination of the same patient and seven because they did not reach minimum quality. The final sample was 141 x-rays.

MAIN MEASUREMENTS

The observers filled in independently, for each examination, an 11-item protocol on the technical quality of the images. The kappa index between pairs of observers was calculated for each item, as was the overall kappa index.

RESULTS

96% of the examinations were conducted with large x-ray plates (3543). There was only acceptable or good concordance between the 3 observers in 2 questions (kappa, 0.559-0.858). In 5 questions concordance was homogeneously low (kappa, 0.034-0.375). In some questions there was a strong discrepancy between the appraisal of the expert and that of the 2 other observers.

CONCLUSIONS

General concordance can be considered low, although it is somewhat greater between the radiologist and the family doctor than between either of these and the expert. This poses the need to improve professional training in evaluation of the technical quality of images. In addition, there was unnecessary expense in large-size x-ray plates.