Physics-aware learning and domain-specific loss design in ophthalmology.

The human cataract, a developing opacification of the human eye lens, currently constitutes the world's most frequent cause for blindness. As a result, cataract surgery has become the most frequently performed ophthalmic surgery in the world. By removing the human lens and replacing it with an artificial intraocular lens (IOL), the optical system of the eye is restored. In order to receive a good refractive result, the IOL specifications, especially the refractive power, have to be determined precisely prior to surgery. In the last years, there has been a body of work to perform this prediction by using biometric information extracted from OCT imaging data, recently also by machine learning (ML) methods. Approaches so far consider only biometric information or physical modelling, but provide no effective combination, while often also neglecting IOL geometry. Additionally, ML on small data sets without sufficient domain coverage can be challenging. To solve these issues, we propose OpticNet, a novel optical refraction network based on an unsupervised, domain-specific loss function that explicitly incorporates physical information into the network. By providing a precise and differentiable light propagation eye model, physical gradients following the eye optics are backpropagated into the network. We further propose a new transfer learning procedure, which allows the unsupervised pre-training on the optical model and fine-tuning of the network on small amounts of surgical patient data. We show that our method outperforms the current state of the art on five OCT-image based data sets, provides better domain coverage within its predictions, and achieves better physical consistency.

Evaluation of intraoperative aphakic eye axial length measurements using swept-source OCT.

PURPOSE: To evaluate intraoperative aphakic eye axial length (AL) measurements using swept-source optical coherence tomography (SS-OCT). SETTING: Hanusch Hospital, Vienna, Austria. DESIGN: Prospective single-center study. METHODS: Patients scheduled for cataract surgery were measured using SS-OCT (IOLMaster 700, Carl Zeiss Meditec AG) to assess the AL. Intraoperatively (intra-OP), SS-OCT measurements were performed with a prototype device (IOLMaster 700 connected to an OPMI Lumera 700 microscope, CZM) at the beginning of cataract surgery furthermore of the aphakic eye and 2 months after surgery. RESULTS: 106 patients were included. Of the 59 eyes of 59 patients, the phakic median AL preoperatively and intra-OP was 23.61 mm ± 0.96 (standard deviation [SD]) and 23.51 mm ± 0.96 (SD), respectively. The absolute median difference was 0.028 ± 0.02 (SD) (P = .049). Median phakic AL intra-OP vs 2 months postoperatively (post-OP) was 23.51 mm ± 0.97 (SD) vs 23.49 mm ± 0.95 (SD). The absolute median difference was 0.049 ± 0.04 (SD) (P = .000). Median AL intra-OP aphakic vs 2 months post-OP pseudophakic was 23.42 mm ± 0.97 (SD) vs 23.42 mm ± 0.97 (SD), respectively. Absolute median difference was 0.038 ± 0.04 (SD) (P = .379). CONCLUSIONS: Intra-OP, SS-OCT technology of the phakic and aphakic eye shows excellent comparability to preoperative and postoperative measurements. This technique allows AL measurements with high precision in cases in which preoperative biometric measurements are not possible.

Feasibility of intrasurgical spectral-domain optical coherence tomography.

PURPOSE: To evaluate the feasibility of intrasurgical spectral-domain optical coherence tomography in a pilot study. METHODS: Using a Carl Zeiss Meditec Cirrus HD-OCT system adapted to the optical pathway of a Zeiss OPMI VISU 200 surgical microscope, 512 × 128 macular cube scans were performed during various steps of microsurgical procedures in 25 cases. The acquired volume data were postprocessed and visualized using a ray-traced three-dimensional display system. RESULTS: The surgical procedures included pars plana vitrectomies for epiretinal membranes (n = 8), macular holes (n = 4), primary rhegmatogenous retinal detachment (n = 1), proliferative diabetic retinopathy (n = 3), silicone oil removal (n = 2), and cataract surgery only (n = 7). It was possible to acquire intraretinal scans with sufficient quality from all patients. Decisions for additional membrane peeling, knowledge about the behavior of the macular hole and the foveal depression during and after membrane removal, information about clinically invisible fluid accumulation under silicone oil or in a clinically diagnosed "macula-on" retinal detachment, and the condition of the fovea immediately after cataract removal could be gained. CONCLUSION: Intrasurgical spectral-domain optical coherence tomography evaluation is feasible using the tested system and may positively influence surgical decisions and techniques resulting in an improved patient outcome.

An evaluation of a transcutaneous and an end-tidal capnometer for noninvasive monitoring of spontaneously breathing patients.

BACKGROUND: Since there is a growing use of analgesia and sedation in spontaneously breathing patients undergoing diagnostic or therapeutic interventions, recommendations by national societies of anesthesiologists call for the application of capnometry during all anesthetic procedures. METHODS: We compared readings from a transcutaneous capnometer (Tosca) and an end-tidal capnometer (Microcap Plus) to P(aCO2) measurements made via arterial-blood-gas analysis. We studied 30 spontaneously breathing patients who were recovering from general anesthesia, and we used Bland Altman analysis to compare the capnometry readings to the arterial-blood-gas values. Expiratory gas samples for end-tidal capnometry were taken either from a conventional face mask or an oral/nasal cannula. RESULTS: The Tosca significantly overestimates P(aCO2) (mean +/- SD difference 5.6 + 3.4 mm Hg). The Microcap Plus significantly underestimates P(aCO2) (mean +/- SD difference -14.1 +/- 7.4 mm Hg). There was no significant difference between the face mask and oral/nasal cannula with regard to collecting end-tidal samples. CONCLUSION: Both the Tosca and Microcap Plus provide just an approximate estimation of P(aCO2). Clinical use of these monitors can not be proposed under actual conditions but will be advantageous after correction of the limiting errors.

The effects of motion artifact and low perfusion on the performance of a new generation of pulse oximeters in volunteers undergoing hypoxemia.

INTRODUCTION: Motion artifact and low perfusion often lead to faulty or absent pulse oximetry readings in clinical practice. OBJECTIVE: Determine the impact of motion artifact and low perfusion on newly introduced pulse oximetry technologies during hypoxemic episodes in healthy volunteers. METHODS: Five different pulse oximeters from 4 manufacturers (the Datex Ohmeda 3900P; the Agilent; the Nellcor N-3000; the Nellcor N-395; and the Schiller OX-1, which is the European version of the Ivy SatGuard 2000 with Masimo SET) were compared with respect to their ability (separated or in combination) to provide accurate readings in the presence of motion artifact and low perfusion. Four of these oximeters represent the latest available oximetry technology, and one (the N-3000) represents a previous generation of oximeters. Oxygen saturation values (S(pO(2))) and pulse rate from the oximeters were recorded during episodes of induced hypoxemia in 10 healthy volunteers. Standardized and repeatable motion artifacts were generated by a motion machine and by having the test subject perform tapping and scratching motions. Perfusion to the finger was reduced by an inflatable balloon impinging on the brachial artery. The pulse oximetry readings from the test oximeters were compared to readings from control pulse oximeters on the unperturbed reference hand. The pulse rates from the test oximeters were compared to the electrocardiographically-measured heart rate. RESULTS: The frequency of faulty readings was increased by increasing motion interference and decreasing perfusion. The S(pO(2)) deviation was within +/- 3% of the reference reading > 95% of the time for all instruments during the control desaturation period in the absence of motion and with normal perfusion. With the combination of motion and low perfusion, the S(pO(2)) error was within +/- 3% less than 62% of the time for all oximeters tested. A significant difference in the frequency of large S(pO(2)) errors was observed only in the direct comparison of the N-395 and N-3000. The N-395 exhibited less frequent S(pO(2)) error exceeding 6% of S(pO(2)) in the combination of the most challenging situations (motion and motion with reduced perfusion). In the same situation the Datex-Ohmeda 3900P and Nellcor N-3000 showed significantly higher pulse rate errors than the other devices (Datex-Ohmeda 3900P 53% of the time and N-3000 37% of the time). CONCLUSIONS: The established model of creating motion artifact and low perfusion is capable of simulating a hierarchy of severe clinical situations. With solely motion or solely reduced perfusion the percentage of errors exceeding +/- 3% of S(pO(2)) increased by 20% and 10%, respectively, compared to the control period. Simultaneous presence of motion and reduced perfusion leads to a relative incidence of > 35% of errors > 3% of S(pO(2)) for the various oximeters. In this situation the N-3000 and the Datex-Ohmeda 3900P exhibited differences between estimated pulse rate and electrocardiographically-measured heart rate > 25 beats/min > 37% of the time.