Introduction Deep Learning (DL) and Artificial Intelligence (AI) have become widespread due to the advanced technologies and availability of digital data. Supervised learning algorithms have shown human-level performance or even better and are better feature extractor-quantifier than unsupervised learning algorithms. To get huge dataset with good quality control, there is a need of an annotation tool with a customizable feature set. This paper evaluates the viability of having an in house annotation tool which works on a smartphone and can be used in a healthcare setting. Methods We developed a smartphone-based grading system to help researchers in grading multiple retinal fundi. The process consisted of designing the flow of user interface (UI) keeping in view feedback from experts. Quantitative and qualitative analysis of change in speed of a grader over time and feature usage statistics was done. The dataset size was approximately 16,000 images with adjudicated labels by a minimum of 2 doctors. Results for an AI model trained on the images graded using this tool and its validation over some public datasets were prepared. Results We created a DL model and analysed its performance for a binary referrable DR Classification task, whether a retinal image has Referrable DR or not. A total of 32 doctors used the tool for minimum of 20 images each. Data analytics suggested significant portability and flexibility of the tool. Grader variability for images was in favour of agreement on images annotated. Number of images used to assess agreement is 550. Mean of 75.9% was seen in agreement. Conclusion Our aim was to make Annotation of Medical imaging easier and to minimize time taken for annotations without quality degradation. The user feedback and feature usage statistics confirm our hypotheses of incorporation of brightness and contrast variations, green channels and zooming add-ons in correlation to certain disease types. Simulation of multiple review cycles and establishing quality control can boost the accuracy of AI models even further. Although our study aims at developing an annotation tool for diagnosing and classifying diabetic retinopathy fundus images but same concept can be used for fundus images of other ocular diseases as well as other streams of medical science such as radiology where image-based diagnostic applications are utilised.

BACKGROUNDPituitary adenoma is a benign and most common tumour of the pituitary gland. It is also the most common parachiasmal tumour and accounts for approximately 10-15% of primary intracranial neoplasms. It has an annual incidence rate of 0.8-8 per 1,00,000 population. Pituitary adenomas are classified as functional and non-functional based on their hormonal activity. Functional adenomas are usually detected earlier due to clinical manifestations produced by excess of hormones.The aim of the study is to analyse visual acuity, visual fields, RNFL thickness and GCIPL thickness on optical coherence tomography (OCT) and to find a correlation between these parameters and tumour volume in patients diagnosed with pituitary adenoma. MATERIALS AND METHODS48 patients diagnosed with pituitary adenoma confirmed by MRI scan underwent complete ophthalmic evaluation (visual acuity, slit-lamp examination, fundus evaluation), perimetry using 30-2 SITA FAST strategy, (Humphrey Field Analyzer; Carl-Zeiss Meditec, Dublin, CA), and OCT of disc (for retinal nerve fibre layer-RNFL thickness) and macula (for ganglion cell-inner plexiform layer (GCIPL) thickness) using Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA) at Bangalore West Lions Super Speciality Eye Hospital, between June 2014 to June 2016. Various parameters like Mean Deviation (MD), Pattern Standard Deviation (PSD) and RNFL and GCIPL thickness on OCT were analysed and correlated with each other. RESULTSMean tumour volume in patients was 12.26 ± 15.8 cm 3 . Most of the patients had visual acuity 6/18 or better. Bitemporal hemianopia was seen in only 5 (12.2%) patients. Superotemporal quadrantanopia, arcuate defects, tubular fields and homonymous hemianopia were the other field defects seen. Total and pattern deviation plot of visual fields correlated well with tumour volume and visual acuity. On visual field analysis, the MD (-8.18 ± 8.65 dB) was depressed compared to the control group (-2.0 ± 1.8 dB), and PSD value (5.76 ± 4.8 dB) was higher than controls (1.9 ± 1.0 dB). However, MD and PSD did not correlate well with tumour volumes. Mean RNFL thickness (85.9 ± 14.5 µm) and mean GCIPL thickness (71.6 ± 17.2 µm), values revealed global thinning in patients when compared with RNFL thickness (92.4 ±7.6 µm) and GCIPL thickness (80.4 ± 4.0 µm) in controls. MD and PSD correlated well with all sectors of RNFL and GCIPL (p value <0.01). CONCLUSIONPSD and GCIPL were concluded to be valuable tools in prognosis of the disease. Our study reinforces the effectiveness of investigations like standard automated perimetry and OCT in prognosticating the neurological disorder like pituitary adenoma and to understand the structural and functional relationship of the disease process. Our study recommends the use of GCIPL thickness evaluation by OCT for patients with unreliable fields or fields not corresponding to the disease progress.

No abstract