Mohanasankar S

Glaucoma is a highly threatening and widespread ocular disease which may lead to permanent loss in vision. One of the important parameters used for Glaucoma screening in the cup-to-disc ratio (CDR), which requires accurate segmentation of optic cup and disc. We explore fully convolutional networks (FCNs) for the task of joint segmentation of optic cup and disc. We propose a novel improved architecture building upon FCNs by using the concept of residual learning. Additionally, we also explore if adversarial training helps in improving the segmentation results. The method does not require any complicated preprocessing techniques for feature enhancement. We learn a mapping between the retinal images and the corresponding segmentation map using fully convolutional and adversarial networks. We perform extensive experiments of various models on a set of 159 images from RIM-ONE database and also do extensive comparison. The proposed method outperforms the state of the art methods on various evaluation metrics for both disc and cup segmentation.

Recently, a training mannequin suitable for ophthalmic blocks has been developed. It provides a new feature of visual feedback to the trainee. A validation of the efficacy of this training system with needle angle visualization, ocular structure proximity, and procedural warning features was conducted in an evaluative study with 41 participants; 21 post graduate trainees and 20 ophthalmic consultants from a tertiary ophthalmic care facility in Chennai, India. The participant's performance was evaluated and analyzed using an appropriate scoring scheme in two sessions with and without visual feedback. The participants were also requested to provide feedback on the anatomical likeness and usage. A two tailed signed rank Wilcoxon test verified the statistical significance of the visual feedback (Pz(4.55, 0.05)= 0.9999, P < 0.001). The mean score of the participants showed an increase 58.86% and 25.5% for graduate trainee and consultants respectively and a mean reduction of 85.41% in the warning indications provided was also substantiated the efficacy of the visual feedback.

Optical tracking systems are extensively used in minimally invasive image-guided surgeries. The efficiency of such a system depends on the precise tracking of surgical tools. The optimal setup of the navigation system and calibration of the tool are the predominant factors that affect the tracking accuracy. We have developed and implemented customized filtering and calibration algorithms on a cost-effective camera, the fusionTrack 500 manufactured by Atracsys. Deployment of these in-house algorithms and protocols have shown a significant increase in the tracking accuracy. Extensive validations were conducted on fusionTrack 500 and benchmarked against NDI Polaris Vega, an industrial standard camera. ASTM phantom was used to validate the tracking accuracy of the navigation system. A Deldrin-based phantom was designed particularly to evaluate the Target Registration Error. Further, a volumetric study was carried out to assess the relative position error between cameras, using a 3D FARO arm CMM. The proposed calibration method implemented on the Atracsys camera shows a 33.7% improvement in the tracking accuracy compared to its native algorithm. Target Registration Error for NDI camera was observed to be 1.2 mm and Atracsys was 0.8 mm, which depicts a 0.4 mm enhancement. Incorporating these algorithms would allow us to effectively integrate cost-effective optical tracking systems into image-guided navigation systems.

Vertebral labelling and segmentation are two fundamental tasks in an automated spine processing pipeline. Reliable and accurate processing of spine images is expected to benefit clinical decision support systems for diagnosis, surgery planning, and population-based analysis of spine and bone health. However, designing automated algorithms for spine processing is challenging predominantly due to considerable variations in anatomy and acquisition protocols and due to a severe shortage of publicly available data. Addressing these limitations, the Large Scale Vertebrae Segmentation Challenge (VERSE) was organised in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2019 and 2020, with a call for algorithms tackling the labelling and segmentation of vertebrae. Two datasets containing a total of 374 multi-detector CT scans from 355 patients were prepared and 4505 vertebrae have individually been annotated at voxel level by a human-machine hybrid algorithm (https://osf.io/nqjyw/, https://osf.io/t98fz/). A total of 25 algorithms were benchmarked on these datasets. In this work, we present the results of this evaluation and further investigate the performance variation at the vertebra level, scan level, and different fields of view. We also evaluate the generalisability of the approaches to an implicit domain shift in data by evaluating the top-performing algorithms of one challenge iteration on data from the other iteration. The principal takeaway from VERSE: the performance of an algorithm in labelling and segmenting a spine scan hinges on its ability to correctly identify vertebrae in cases of rare anatomical variations. The VERSE content and code can be accessed at: https://github.com/anjany/verse.

An electroencephalogram (EEG) signal is currently accepted as a standard for automatic sleep staging. Lately, Near-human accuracy in automated sleep staging has been achievable by Deep Learning (DL) based approaches, enabling multi-fold progress in this area. However, An extensive and expensive clinical setup is required for EEG based sleep staging. Additionally, the EEG setup being obtrusive in nature and requiring an expert for setup adds to the inconvenience of the subject under study, making it adverse in the point of care setting. An unobtrusive and more suitable alternative to EEG is Electrocardiogram (ECG). Unsurprisingly, compared to EEG in sleep staging, its performance remains sub-par. In order to take advantage of both the modalities, transferring knowledge from EEG to ECG is a reasonable approach, ultimately boosting the performance of ECG based sleep staging. Knowledge Distillation (KD) is a promising notion in DL that shares knowledge from a superior performing but usually more complex teacher model to an inferior but compact student model. Building upon this concept, a cross-modality KD framework assisting features learned through models trained on EEG to improve ECG-based sleep staging performance is proposed. Additionally, to better understand the distillation approach, extensive experimentation on the independent modules of the proposed model was conducted. Montreal Archive of Sleep Studies (MASS) dataset consisting of 200 subjects was utilized for this study. The results from the proposed model for weighted-F1-score in 3-class and 4-class sleep staging showed a 13.40 % and 14.30 % improvement, respectively. This study demonstrates the feasibility of KD for single-channel ECG based sleep staging's performance enhancement in 3-class (W-R-N) and 4-class (W-R-L-D) classification.

It is an arduous task to perform an Image Quality Assessment (IQA) for all real-time imaging systems like endoscopy systems which have no reference images for comparison. That being a challenge in itself, the images must be assessed subjectively rather than objectively so that the assessments comply with human perception. ImageLab, a complete IQA solution developed in-house has image capture and illumination capabilities integrated with custom test charts and algorithms to provide a detailed analysis for individual image parameters. Though individual parameter analysis is the primary focus of the IQA tool, a technique to provide a score for image quality based on the interdependency of the image parameters is also integrated in the tool. Specifically, the dependency of visual noise and dynamic range on sharpness and resolution has been highlighted in this paper. Eight levels of sharpness were chosen and for every level, image quality in terms of visual noise and dynamic range were computed. The perceived visual noise levels which is part of subjective analysis were factored in the dynamic range calculations and the final image quality score was derived. These detailed measurements would aid in the development of endoscopy systems and the much simpler score-based technique would aid the technicians during maintenance and calibration check before endoscopy examination.

Professionals having desk jobs have reported increased levels of stress in recent times. A major reason for this can be attributed to sitting in poor postures for prolonged period of time. Maintaining such postures, especially under high mental workloads, may lead to reduced Heart Rate Variability (HRV), leading to increased stress levels, thereby increasing the risk of cardiovascular health disorders such as heart attacks, strokes, etc. Therefore, this study investigated the influence of both right and poor posture on HRV during high mental workload conditions. We recorded the Heart Rate (HR), using chest based wearable, of ten healthy subjects under controlled environmental conditions. All the measurements were obtained with each subject maintaining different postures such as High Power (HP) pose, Low Power (LP) pose, High Power pose under mental load conditions (HPm), and Low Power pose under mental load conditions (LPm). HRV indices like time domain and frequency domain for the above postures were computed. We observed that eight subjects showed higher HRV during HP pose compared to LP pose. Likewise, the subjects showed higher HRV during HPm pose compared to LPm pose. We studied the impact of postural change on HRV through statistical t-paired test and observed a significant difference (p < 0. 05) in the HRV indices across subjects for different postures. Hence, this study demonstrated the positive impact of the right posture on HRV of individuals having desk jobs.

Automation of life sciences and pharmaceutical laboratories is high on demand in the present day, especially with the emergence of high throughput screening and drug discovery. Lab automation provides reliable and efficient solutions, limiting reagent amounts with increased throughput and reduced costs. In this work, we present an affordable, intelligent, sub-microlitre resolution automatic pipetting system. This system employs a differential pressure sensor to detect most frequently occurring pipetting errors in automatic liquid handling workstations, which include no liquid sample present in the source well, pipette tip not attached to the holder, clogging of the tip, aspiration of air bubbles or foam, insufficient liquid in the source well, improper dispensing of the samples and tip not intact with the holder. This also determines liquid volumes aspirated into the pipette tip, providing a self-validating mechanism for the system, thereby reducing human intervention and improving performance efficiency. A prototype system was developed in-house and gravimetric analysis was carried out to determine system performance. Repeatability coefficient of variation for the designed system was found to be below 5.5% for the volume range of 10 to 200 microlitres, while the error obtained was observed to be less than 1.6%. Experimental tests conducted could flawlessly detect mentioned pipetting abnormalities through the algorithm designed.

An arterial compliance, dual-element photoplethysmograph probe for local pulse wave velocity (PWV) measurement was developed. Initially, the experimental validation study was performed on 25 young volunteers (age =24.5 ± 4 years). Local PWV was assessed from a small section (23 mm) of the carotid artery. The prototype device demonstrated its capability of measuring reliable, repeatable, and reproducible carotid local PWV. Further, in 15 healthy male volunteers (age = 22.25 ± 3.5 years), carotid local PWV and brachial blood pressure (BP) were continuously recorded during their postexercise recovery period. Local PWV followed the changes in arterial BP parameters. The group average correlation coefficients (r) of local PWV versus BP parameters were between 0.772 ± 0.033 and 0.934 ± 0.028. In a population of 50 patients (normotensive and hypertensive) aged 24-80 years, local PWV-BP correlations were investigated. Local PWV tended to follow the diastolic BP (DBP; r = 0.82) and mean arterial pressure (r = 0.83) better than systolic BP (SBP; r = 0.69). It was significantly inferior in tracking pulse pressure values (r = 0.35). Cuffless estimation of arterial pressure was also performed on the same patients using measured carotid local PWV with best-case calibrations. Local PWV yielded good DBP prediction than SBP prediction. Statistically, significant correlation (r = 0.79) and a root-mean-square error of 5.26 mmHg versus reference brachial DBP were achieved. The introduced technique has a potential for short- or long-term noninvasive, cuffless monitoring of BP parameters from superficial arteries.

Palpation is the process of using one's hand to perform a physical examination. Ophthalmic anaesthetists palpate the orbital bone around the eye, to locate anatomical markers to help them guide the needle safely for regional needle block into the orbital cavity. The anaesthetists are provided very little training on palpation procedure due to the lack of a suitable training system. Inadvisable palpation can cause damage to the soft tissue if the applied force is more than required. There is a necessity to provide ophthalmic anaesthetists with the relevant knowledge about palpation and the force exerted during this procedure. In this paper, a load cell based palpation sensor which can be integrated with recently developed ophthalmic anaesthesia training systems with a virtual instrument environment that is capable of mimicking the skin reaction to applied palpation with the variable warning threshold is proposed.