Background/Aims The kidneys and retina are highly vascularized organs that frequently exhibit shared pathologies, with nephropathy often associated with retinopathy. Previous studies have successfully predicted estimated glomerular filtration rates (eGFRs) using fundus photographs. We evaluated the performance of the Modification of Diet in Renal Disease (MDRD) and Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formulas in eGFR prediction. Methods We enrolled patients with fundus photographs and corresponding creatinine measurements taken on the same date. One photograph per eye was randomly selected, resulting in a final dataset of 45,108 patients (88,260 photographs). Data including sex, age, and blood creatinine levels were collected for eGFR calculation using the MDRD and CKD-EPI formulas. EfficientNet B3 models were used to predict each parameter. Results Deep neural network models accurately predicted age and sex using fundus photographs. Sex was identified as a confounding variable in creatinine prediction. The MDRD formula was more susceptible to this confounding effect than the CKD-EPI formula. Notably, the CKD-EPI formula demonstrated superior performance compared to the MDRD formula (area under the curve 0.864 vs. 0.802). Conclusions Fundus photographs are a valuable tool for screening renal function using deep neural network models, demonstrating the role of noninvasive imaging in medical diagnostics. However, these models are susceptible to the influence of sex, a potential confounding factor. The CKD-EPI formula, less susceptible to sex bias, is recommended to obtain more reliable results.
Breast cancer accounts for one in four new malignant tumors in women, and misdiagnosis can lead to severe consequences, including delayed treatment. Among patients classified with a BI-RADS 3 rating, the risk of very early-stage malignancy remains over 2%. However, due to the benign imaging characteristics of these lesions, radiologists often recommend follow-up rather than immediate biopsy, potentially missing critical early interventions. This study aims to develop a deep learning (DL) model to accurately identify very early-stage malignancies in BI-RADS 3 lesions using ultrasound (US) images, thereby improving diagnostic precision and clinical decision-making. A total of 852 lesions (256 malignant and 596 benign) from 685 patients who underwent biopsies or 3-year follow-up were collected by Southwest Hospital (SW) and Tangshan People’s Hospital (TS) to develop and validate a deep learning model based on a novel transfer learning method. To further evaluate the performance of the model, six radiologists independently reviewed the external testing set on a web-based rating platform. The proposed model achieved an area under the receiver operating characteristic curve (AUC), sensitivity, and specificity of 0.880, 0.786, and 0.833 in predicting BI-RADS 3 malignant lesions in the internal testing set. The proposed transfer learning method improves the clinical AUC of predicting BI-RADS 3 malignancy from 0.721 to 0.880. In the external testing set, the model achieved AUC, sensitivity, and specificity of 0.910, 0.875, and 0.786 and outperformed the radiologists with an average AUC of 0.653 (<i>p</i> = 0.021). The DL model could detect very early-stage malignancy of BI-RADS 3 lesions in US images and had higher diagnostic capability compared with experienced radiologists.
Photography, Computer applications to medicine. Medical informatics
Nicole Aberle, Torkild Bakken, Luis Martell
et al.
Jellyfish form irregular and seemingly unpredictable blooms that can be the result of redistribution/aggregation events or peaks in population growth. Such blooms can affect ecosystem structure and stability due to the role of jellyfish as top predators of fish larvae and eggs and as competitors of fish preying on the same zooplankton resources. Factors leading to jellyfish bloom formation have received a lot of attention during the past decades. However, factors causing blooms to collapse are less studied. The helmet jellyfish Periphylla periphylla is a bloom-forming jellyfish species that shows mass occurrences in several Norwegian fjords with substantial socioeconomic implications. The success of P. periphylla to form massive blooms is attributed to its longevity, the lack of natural predators and its holoplanktonic life cycle with a continuous reproduction throughout the year. In Trondheimsfjorden, central Norway, P. periphylla has established large populations over the last decades. However, population estimates with a high spatiotemporal resolution are scarce and the regulating mechanisms affecting P. periphylla dynamics remain largely unknown. Using in-situ imaging techniques during pelagic dives with a remotely operated vehicle (ROV), enabled insights on the potential role of parasites as bloom-controllers. ROV footage provided unique information on high levels of prevalence and intensity of parasitic amphipods in P. periphylla. In parallel, these parasitic associations were documented by underwater photography in the same area and season. The combination of non-invasive imaging techniques allowed estimates on the degree of parasite infestation and on how parasitism can affect the condition of jellyfish populations thus eventually causing blooms to collapse. We suggest taking investigations on parasite-host interactions and the role of parasitism as a population control mechanism into the spotlight. Future research in this field will benefit from using non-invasive imaging tools to study parasite-host interactions and animal behaviour in-situ.
Corn is an important economic and food crop, and the corn threshing process is an important link in the processing of corn, but the damage rate in the threshing process has always been a problem, causing difficulties in subsequent processing and storage. To address the high damage rate in corn ear threshing, a texture analyzer was used to measure the fracture force of Boyun 88 and Zhengdan 958 corn varieties in the triaxial direction, and a CT scanning imaging system was used to analyze the connection mode between the carpopodium and the corn cob. The connection between the carpopodium and corn cob, as well as the fracture process of the carpopodium, was simulated. Finally, high-speed photography was used to study the corn ear threshing process. The results indicated that the fracture force of the carpopodium under radial tension was significantly greater than that under axial and tangential shear. Additionally, the simulated fracture stress value of the carpopodium exceeded its actual fracture stress value. Under radial stress, the fracture force between the carpopodium and corn cob exhibited more uniformity on the contact surface. When a tangential load was applied, it was observed that the force chain shifted and dissipated along the axis during corn kernel extrusion. High-speed photography on a discrete test bench revealed that corn kernel dispersion, extrusion, and force transfer facilitated the movement and migration of surrounding kernels, with the force transfer process resembling a “trapezoid”. This study offers theoretical guidance for corn threshing with low damage and an analysis of the threshing process.
Joseph Dias, Puvan Tharmanathan, Catherine Arundel
et al.
Background Dupuytren’s contracture is caused by nodules and cords which pull the fingers towards the palm of the hand. Treatments include limited fasciectomy surgery, collagenase injection and needle fasciotomy. There is limited evidence comparing limited fasciectomy with collagenase injection. Objectives To compare whether collagenase injection is not inferior to limited fasciectomy when treating Dupuytren’s contracture. Design Pragmatic, two-arm, unblinded, randomised controlled non-inferiority trial with a cost-effectiveness evaluation and nested qualitative and photographic substudies. Setting Thirty-one National Health Service hospitals in England and Scotland. Participants Patients with Dupuytren’s contracture of ≥ 30 degrees who had not received previous treatment in the same digit. Interventions Collagenase injection with manipulation 1–7 days later was compared with limited fasciectomy. Main outcome measures The primary outcome was the Patient Evaluation Measure score, with 1 year after treatment serving as the primary end point. A difference of 6 points in the primary end point was used as the non-inferiority margin. Secondary outcomes included: Unité Rhumatologique des Affections de la Main scale; Michigan Hand Outcomes Questionnaire; recurrence; extension deficit and total active movement; further care/re-intervention; complications; quality-adjusted life-year; resource use; and time to function recovery. Randomisation and blinding Online central randomisation, stratified by the most affected joint, and with variable block sizes allocates participants 1 : 1 to collagenase or limited fasciectomy. Participants and clinicians were not blind to treatment allocation. Results Between 31 July 2017 and 28 September 2021, 672 participants were recruited (n = 336 per group), of which 599 participants contributed to the primary outcome analysis (n = 285 limited fasciectomy; n = 314 collagenase). At 1 year (primary end point) there was little evidence to support rejection of the hypothesis that collagenase is inferior to limited fasciectomy. The difference in Patient Evaluation Measure score at 1 year was 5.95 (95% confidence interval 3.12 to 8.77; p = 0.49), increasing to 7.18 (95% confidence interval 4.18 to 10.88) at 2 years. The collagenase group had more complications (n = 267, 0.82 per participant) than the limited fasciectomy group (n = 177, 0.60 per participant), but limited fasciectomy participants had a greater proportion of ‘moderate’/‘severe’ complications (5% vs. 2%). At least 54 participants (15.7%) had contracture recurrence and there was weak evidence suggesting that collagenase participants recurred more often than limited fasciectomy participants (odds ratio 1.39, 95% confidence interval 0.74 to 2.63). At 1 year, collagenase had an insignificantly worse quality-adjusted life-year gain (−0.003, 95% confidence interval −0.006 to 0.0004) and a significant cost saving (−£1090, 95% confidence interval −£1139 to −£1042) than limited fasciectomy with the probability of collagenase being cost-effective exceeding 99% at willingness to pay thresholds of £20,000–£30,000 per quality-adjusted life-year. At 2 years, collagenase was both significantly less effective (−0.048, 95% confidence interval −0.055 to −0.040) and less costly (−£1212, 95% confidence interval −£1276 to −£1147). The probability of collagenase being cost-effective was 72% at the £20,000 threshold but limited fasciectomy became the optimal treatment at thresholds over £25,488. The Markov model found the probability of collagenase being cost-effective at the lifetime horizon dropped below 22% at thresholds over £20,000. Semistructured qualitative interviews found that those treated with collagenase considered the outcome to be acceptable, though not perfect. The photography substudy found poor agreement between goniometry and both participant and clinician taken photographs, even after accounting for systematic differences from each method. Limitations Impacts of the COVID-19 pandemic resulted in longer waits for Dupuytren’s contracture treatment, meaning some participants could not be followed up for 2 years. This resulted in potential underestimation of Dupuytren’s contracture recurrence and/or re-intervention rates, which may particularly have impacted the clinical effectiveness and long-term Markov model findings. Conclusions Among adults with Dupuytren’s contracture, collagenase delivered in an outpatient setting is less effective but more cost-saving than limited fasciectomy. Further research is required to establish the longer-term implications of both treatments. Future work Recurrence and re-intervention usually occur after 1 year, and therefore follow-up to 5 years or more could resolve whether the differences observed in the Dupuytren’s interventions surgery versus collagenase trial to 2 years worsen. Study registration Current Controlled Trials ISRCTN18254597. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 15/102/04) and is published in full in Health Technology Assessment; Vol. 28, No. 78. See the NIHR Funding and Awards website for further award information.
Plain language summary Dupuytren’s contracture happens when fibrous tissue builds up and over time bends the finger(s) into the palm, causing problems with hand function. To treat this, surgery is usually used to straighten the finger. A less intrusive alternative is an injection (collagenase), which softens the tissue after which the finger is moved to straighten it. The Dupuytren's interventions surgery versus collagenase trial recruited 672 patients who were equally and randomly assigned to have either surgery or collagenase injection. The study assessed whether the injection was as good and as safe as surgery at straightening the finger and how long the finger remained straightened. For up to 2 years after treatment, the participant’s hand function and general health were assessed. Some participants provided photographs to monitor changes to the finger, and some were asked about their experiences of Dupuytren’s contracture and treatments. We found: Hand health improved following both treatments. Initially, the injection treatment improved hand health more than surgery. However, by 1 year, surgery improved hand health more than the injection treatment. Recovery of hand function was quicker for participants who received the injection; however, they were more likely to need further treatment (i.e. further care and/or re-intervention). Participants said that the less positive longer-term outcome was acceptable for a better treatment experience. For both treatments, interviews found that participants were happy with the hand improvement they experienced at 3 months after treatment. More than half of participants had no complications, moderate or severe complications were rare, and participants who had surgery had more of these. The injection was cheaper but less effective than surgery at 1 year and was considered good value for money. However, by 2 years surgery became the better option due to its greater improvement in health benefits. Participant-taken photographs can help monitor Dupuytren’s contracture but do not give the same results as measurements taken in a clinic.
Scientific summary Background Dupuytren’s disease affects over 2 million UK adults. Cords pull the fingers down towards the palm. This interferes with hand function and dexterity, impacting on quality of life. Current treatments to remove, dissolve or break the cords include surgical correction [limited fasciectomy (LF)], collagenase injection (an enzyme injected into the cord), and percutaneous needle fasciotomy (a needle is used to puncture, weaken and cut the cord). None of these treatments cure the tendency to develop Dupuytren’s contracture (DC) and so the cords and contracture can recur over time. Collagenase has some benefits over LF surgery including shorter recovery and no dependence on operating theatre availability for delivery of the intervention. There is, however, limited robust evidence comparing surgical correction and collagenase injection in terms of clinical effectiveness, cost-effectiveness, and in terms of patient’s experiences and preferences. Objectives The primary objective was to compare whether collagenase injection is not inferior to LF in the treatment of DC. Secondary objectives included investigation of recurrence at 1 and 2 years after treatment and cost-effectiveness. A qualitative substudy explored patients’ views of collagenase and LF, and a photography substudy investigated whether measurements of extension and flexion made on photographs taken by patients reflect goniometric measurements to assess recurrence. Methods Design The Dupytren's interventions surgery vs collagenase (DISC) trial was a multicentre, pragmatic, parallel two-arm randomised controlled non-inferiority trial with a cost-effectiveness evaluation, and nested qualitative and photography substudies. Participants were randomised on an equal basis to receive either of the two treatment options via a remote randomisation service. Randomisation was blocked, with randomly varying block sizes, and stratified by reference (worst-affected) joint [metacarpophalangeal (MCP) joint or proximal interphalangeal (PIP) joint]. Participants were followed up at 3 months, 6 months, 1 year and 2 years after treatment. Data collection included joint measurements and photography at baseline, and all follow-up time points. Setting Trial recruitment was undertaken in 31 NHS hand units across England and Scotland between June 2017 and September 2021. Participants Patients aged 18 years and over with a discrete, palpable Dupuytren’s cord causing contracture of ≥ 30 degrees and who were appropriate for both study treatments, were eligible for inclusion. Patients were excluded if they had severe contractures (> 135 degrees); had received treatment to the study digit; had other pre-existing disorders affecting hand function; had contraindications to collagenase; had a coagulation disorder; were female and pregnant or breastfeeding; had participated in a study involving another investigational medicinal product within 12 weeks or had another disease or disorder which would put them at risk if participating. Interventions The intervention was collagenase Clostridium histolyticum injection, supplied through routine NHS stocks. Collagenase was injected as three aliquots at set anatomical points in line with the current approved summary of product characteristics. After an interval of 1–7 days, participants returned to the clinic, where under local anaesthetic the cord was snapped to correct the contracture. The control group received LF surgery to remove the diseased nodules and cord to correct the contracture. Participants were followed up at routine wound check appointments following intervention. Outcomes The primary outcome was the Patient Evaluation Measure (PEM) score (0–100 with higher scores indicating worse outcome) at 1 year after treatment. The PEM was also completed at 3 months, 6 months and 2 years after treatment. Secondary outcomes included the Unité Rhumatologique des Affections de la Main (URAM) scale, Michigan Hand Outcomes Questionnaire (MHQ), recurrence, extension deficit and total active movement, complications, further treatments (including further care and/or re-intervention), health-related quality of life [EuroQol-5 Dimensions, five-level version (EQ-5D-5L)], resource use, time to recovery of function (using a single assessment numeric evaluation measure) and overall hand assessment. All outcomes were collected at 3 months, 6 months, 1 year and 2 years. The PEM was also recollected immediately prior to treatment delivery, and the time to recover function and quality of life were also collected at 2 and 6 weeks after treatment. Outcomes were collected primarily in hospital clinics, with some participants being followed up for postal, telephone, or video data collection during the COVID-19 pandemic. The qualitative substudy explored participants experiences of DC and treatments. The photography substudy explored the agreement between measurements obtained using a goniometer and photographs taken by participants at home, to determine whether the two methods of measurement might feasibly be used interchangeably. Results Clinical effectiveness In total 672 participants (64.6%) were recruited and randomised; 336 to receive collagenase injection and 336 to receive LF. Baseline characteristics were similar across groups. Of the 672 randomised participants, 621 (92.4%) received treatment as part of the trial. Cross-over was limited: one participant (0.3%) allocated to collagenase received LF; seven participants allocated to LF received collagenase (2.1%). On average participants received collagenase by 12.1 weeks [standard deviation (SD) 13.7] and LF in 17.7 weeks (SD 16.5) after randomisation. Most participants (n = 315, 95.2%) had just one digit treated. No participants required an unplanned inpatient admission following treatment and 62.0% (n = 201) collagenase participants and 78.3% (n = 224) LF participants had full correction following treatment. At 1 year (primary time-point) the difference in PEM scores showed that collagenase was inferior to LF; difference 5.95 [95% confidence interval (CI) 3.12 to 8.77; p = 0.49]. The benefit of LF over collagenase continued to increase to 2 years (7.18, 95% CI 4.18 to 10.88; p = 0.82). There were no material changes in these results for any of the sensitivity or additional analyses undertaken. The primary analysis therefore shows that there is little evidence to support rejection of the hypothesis that collagenase is inferior to LF at 1 and 2 years post treatment. Indeed, the observed data are highly compatible with LF being superior to collagenase with regard to the primary outcome measure at both these time points. Patient Evaluation Measure overall assessment scores corresponded with the primary outcome analyses and participants in both groups reported positive experiences of treatment. The estimated difference in URAM scores followed those of PEM, increasing in favour of LF over time from 3 months (0.82, 95% CI −0.21 to 1.84; p = 0.12) to 5.37 (95% CI 3.85 to 6.88; p ≤ 0.00005) at 2 years. At 1 year MHQ scores were higher (better) in the LF group (1 year: −4.69, 95% CI −7.27 to −2.12; p = 0.0004) and this continued at 2 years (2 years: −6.71, 95% CI −9.60 to −3.82; p ≤ 0.00005). Return to function was better in the short term for the collagenase group (week 2: 14.93, 95% CI 11.66 to 18.19; p ≤ 0.00005; 6 weeks: 5.00, 95% CI 2.29 to 7.70; p = 0.003) but by 1 year function was superior after LF (−4.93, 95% CI −7.63 to −2.22; p = 0.0004). At 1 year participants who received LF were more likely to respond as being ‘cured’ or ‘much better’ than participants who received collagenase [odds ratio (OR) 3.01, 95% 2.15 to 4.23; p ≤ 0.00005]. Passive extension deficit was similar between the groups at baseline (mean: 45.8°; SD 17.0). Following collagenase treatment, extension deficit seemed to be worse at all time points ranging from a difference of 5.73° (95% CI 2.88 to 8.59; p = 0.0001) at 3 months to 10.10° (95% CI 6.46 to 13.73; p ≤ 0.00005) at 1 year and increasing again up to 2 years. Results when imputed data were included were similar. Increases in reference joint passive range of movement (RoM) were similar between the two groups following treatment. However, from 6 months there was strong evidence that collagenase resulted in poorer passive RoM (−7.42°, 95% CI −11.54 to −3.29; p = 0.0004) and this difference increased further over time. Measurements of active extension deficit were similar between the two groups at baseline (mean: 51.9°, SD 16.1). Like passive extension deficit, active extension deficit was worse following collagenase treatment at all time points, ranging from a difference at 3 months of 5.57° (95% CI 3.02 to 8.12; p ≤ 0.00005) to 11.52° (95% CI 8.13 to 14.91; p < 0.00005) at 1 year and increasing again up to 2 years. Results when imputed data were included were similar. Increases in active RoM of the reference joint were similar between the two groups following treatment. However, from 6 months there was strong evidence that collagenase resulted in poorer active RoM (−8.37°, 95% CI −11.99 to −4.75; p ≤ 0.00005). Again, this difference increased further over time. In total 54 participants (15.7%) experienced recurrence of DC. There was weak evidence to suggest that following collagenase treatment participants were more likely to experience recurrence compared to participants who received LF (OR 1.39, 95% CI 0.74 to 2.63; p = 0.31). There were 267 complications (0.82 per participant) reported for the collagenase group, compared to 177 complications (0.60 per participant) reported for the LF group. Participants in the LF group experienced a higher proportion of ‘moderate’ or ‘severe’ complications (5% vs. 2%). In the first year following intervention, most participants did not require re-intervention (n = 399, 64.3%), which dropped to 47.7% by 2 years. By 2 years, 10% of collagenase participants had re-intervention compared to 2.5% of LF participants. Cost-effectiveness The mean cost of surgery was estimated to be £2510 (SD £818) per participant compared to £1008 (SD £94) for the collagenase group. The overall mean healthcare cost was slightly lower in the collagenase group compared to the LF group at 2 years (mean difference: −£28, 95% CI −£87 to £30). Baseline utility scores (EQ-5D-5L) were slightly higher in the LF group (mean 0.794, SD 0.170) compared to the collagenase group (mean 0.791, SD 0.174), but this was not statistically significant (95% CI −0.029 to 0.024). For both groups, utility scores decreased immediately following treatment but by 3 months had reverted to baseline levels. The mean difference between groups at 2 years was −0.044 (95% CI −0.077 to −0.010). After adjustment for baseline costs and utilities, participants who received collagenase showed a statistically insignificant decrease in quality-adjusted life-year (QALY) gains at 1 year (−0.003, 95% CI −0.006 to 0.0004) and a reduced cost (−£1090, 95% CI −£1139 to −£1042) compared to LF participants. The probability of collagenase being cost-effective was over 99% for both willingness-to-pay thresholds of £20,000 and £30,000 per QALY at 1 year and this finding was robust for the sensitivity analyses conducted. At 2 years collagenase continued to be both significantly less costly (−£1212, 95% CI −£1276 to −£1147) and less effective (−0.048, 95% CI −0.055 to −0.040). The probability of collagenase being cost-effective was 72% at the £20,000 threshold and 37% at the £30,000 threshold. The longer-term Markov model indicated that collagenase became less cost-effective than LF at the lifetime horizon, the probability of collagenase being cost-effective ranged from 22% to 16%. Qualitative Semistructured qualitative interviews were conducted with 45 patients, resulting in four core topics: Lived experience; knowledge; experience; and looking to the future. Participants reported living for extended periods with DC and seeking medical advice only when impacted by the difficulty in doing tasks or appearance of the hand. Most participants reported improvement in their contracture and function; some treated with collagenase noted that while the outcome was not perfect, it was acceptable. More participants treated with collagenase reported preferring this in the future compared to LF participants preferring the same intervention again. Photography substudy The difference between goniometric measurements and participant-taken photographs for active extension deficit was −9.7° (SD 16.2) for MCP, 8.0° (SD 15.1) for PIP and 5° (SD 9.5) for distal interphalangeal (DIP) joints. The limits of agreement were approximately ± 30° for MCP from ± 12° to ± 30° for PIP and ± 18° for DIP joints. For flexion, differences were −0.8° (SD 19.3) for MCP, −1.6° (SD 14.5) for PIP and −2.7° (SD 13.5) for DIP joints. Limits of agreement were approximately ± 36° for MCP, ± 20° for PIP, and a range of ± 33° to ± 24° for DIP joints. Conclusions In adults with moderate DC, collagenase, when delivered in an outpatient setting, proves to be significantly cost-saving compared to LF throughout the trial. While collagenase demonstrates comparable QALY gains to LF at 1 year, its effectiveness is significantly lower at 2 years. This leads to a changing cost-effectiveness profile over time, with collagenase being highly likely cost-effective at 1 year. However, the probability of its cost-effectiveness declines at 2 years. The Markov model results indicate that the likelihood of collagenase being considered cost-effective compared to LF at the lifetime horizon falls below 22% at thresholds of £20,000/QALY and above. The DISC trial followed participants for up to 2 years after treatment and therefore further research is required to better understand the longer-term trajectories for patients following initial contracture correction. Implications for health care The results from the DISC trial provide strong indicators for the planning of care of DC patients in the UK. The comprehensive nature of the clinical and cost-effectiveness data provides the opportunity for the UK National Institute for Health and Care Excellence to update its recommendation on the treatment options for DC. Of relevance will be how to situate the use of collagenase if it is reintroduced for use in the NHS. The role of primary care services in ensuring timely first diagnosis and referral of patients with DC needs to be strengthened. The results of the DISC trial provide a basis to engage further with primary care providers in relation to this. The DISC photography substudy provides an indication of how patient-taken photographs can complement clinic measurements if processes are streamlined further. Further investigation will be key in establishing remote assessment and follow-up for DC patients but noting that clinic measurements remain necessary for final decisions on required care. Recommendations for future research Follow-up to 5 years or more would establish the evolution of differences observed at 2 years, particularly in relation to recurrence and re-intervention, which usually occurs after 1 year. Also, the data collection in the DISC trial has been used as the basis for planning the data collection for the ongoing HAND-2 trial [NIHR: 127393; ISRCTN: 18254597], which will allow for a network meta-analysis of all key interventions for DC. The results from the qualitative substudy provide direction on planning further research to understand behavioural trends that influence a patient’s decision to seek care and return to care after initial intervention. Study registration Current Controlled Trials ISRCTN18254597. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 15/102/04) and is published in full in Health Technology Assessment; Vol. 28, No. 78. See the NIHR Funding and Awards website for further award information.
Abstract Accurately segmenting medical images is a critical step in clinical diagnosis and developing patient‐specific treatment plans. While supervised learning algorithms have achieved excellent performance in this area, they require a large amount of annotated data, which is often time‐consuming and difficult to obtain. As a result, semi‐supervised learning (SSL) has gained attention as it has the potential to alleviate this challenge by using not only limited labelled data but also a large amount of unlabelled data. A common approach in SSL is to filter out high‐entropy features and use the low‐entropy part to compute unsupervised loss. However, it is believed that the high‐entropy part is also beneficial for model training, and discarding it can lead to information loss. To address this issue, a simple yet efficient contrastive learning approach is proposed in this work for semi‐supervised medical image segmentation, called Entropy‐Guided Contrastive Learning Segmentation Network (EGCL‐Net). The proposed method separates the low‐entropy and high‐entropy features via the average of predictions, using contrastive loss to pull the intra‐class entropy representation distance close and push the inter‐class entropy representation distance away. Extensive experiments on the automated cardiac diagnosis challenge dataset, COVID‐19, and BraTS2019 datasets showed that: (1) EGCL‐Net can significantly improve performance by utilizing high‐entropy representation, and (2) the authors’ EGCL‐Net outperforms recent state‐of‐the‐art semi‐supervised methods in both qualitative and quantitative evaluations.
Deqin County in Yunnan Province is among the most severely affected regions in China by debris flow disasters. The Yizhong River in Deqin County has witnessed numerous large-scale debris flow disasters, causing significant damage and substantial economic losses to residential areas and the G214 national road. To elucidate the range of hazard zones and initiation mechanisms of debris flow disaster triggered by potential sources in the upstream Yizhong River under conditions of heavy rainfall and earthquakes, this study conducted field investigations and causal analyses. High-precision Digital Elevation Model (DEM) data derived from close-range UAV aerial photography were utilized as topographic data. The RAMMS software simulated a debris flow of 16.05×104 m3 under heavy rain and earthquake conditions. Two hazardous zones within the Yizhong River Basin were delineated, and the disaster initiation mode of debris flow in Yizhong River was expounded. The results show that the debris flow in Yizhong River belongs to the gully-type viscous debris flow typical of plateau mountainous regions, characterized by large scale, high frequency, and severe impact. Its disaster mechanism is summarized as a gully and valley disaster chain involving high-altitude landslide, debris flow, dammed lake, and flood breach. Risk zone I is located in the area from G214 national road to Dewei Road, while risk zone II is in the gully mouth area prone to accumulation and blockage, presenting high risk. During debris flow movement, the maximum flow velocity reached 23.93 m/s, maximum impact force was 1000 kPa, maximum accumulation depth was 9.33 m, and the maximum single outburst volume of debris flow was approximately 80000 m3, with a danger area of about 0.31 km2. The research results provide a scientific basis for debris flow control projects in Yizhong River and are of practical significance for improving the comprehensive prevention and control of geological hazards in Deqin County.
This article emanates from a geospatial database of over 600 premieres of the Cines company’s Quo Vadis? (1913), an eight-reel film distributed by George Kleine, and nearly 250 premieres of the Quo Vadis Film Company’s Quo Vadis? (1913), a three-reel film of ambiguous origins distributed by Paul De Outo. By mapping local premieres of both films across the United States from 1913 through 1916, the data show with spatiotemporal precision the spread of Quo Vadis? as one of cinema’s early blockbuster titles. Yet within this national phenomenon, the two films’ footprints reveal differing cultural geographies served by competing efforts to feature Quo Vadis? using alternative practices of distribution and exhibition. The study finds that Quo Vadis? played a more complex role mediating the rise of features than is yet known, serving rival modes of cinema where longer, more expensive films were celebrated but also contested.
Purpose: There is a clinical need for a cost-effective, reliable, easy-to-use, and portable retinal photography. The use of smartphone fundus photography for documentation of retinal changes in resource-limited settings, where retinal imaging was not previously possible, is studied here. The introduction of smartphone-based retinal imaging has resulted in the increase in available technologies for fundus photography. On account of the cost, fundus cameras are not readily available in ophthalmic practice in developing countries. Because smartphones are readily available, easy to use, and also portable, they present a low-cost alternative method in resource-limited settings. The aim is to explore the use of smartphones (iphones) for retinal imaging in resource-limited settings. Methods: A smartphone (iphone) was used to acquire retinal images with the use of +20 D lens in patients with dilated pupils by activating the video mode of the camera. Results: Clear retinal images were obtained in different clinical conditions in adults and children, including branch retinal vein occlusion with fibro-vascular proliferation, choroidal neo-vascular membranes, presumed ocular toxoplasmosis, diabetic retinopathy, retinoblastoma, ocular albinism, and hypertensive retinopathy. Conclusion: New inexpensive, portable, easy-to-operate cameras have revolutionized retinal imaging and screening programs and play an innovative role in research, education, and information sharing.
Purpose: X-linked juvenile retinoschisis (XLRS), caused by mutations in the RS1 gene, is an X-linked recessive inherited disease that typically involves both eyes in the first 2 decades of life. Recently, the phenotype heterogeneity of this condition has drawn increasing attention. We reported various phenotypes caused by RS1 gene mutations in eleven patients from ten Chinese families.Methods: Data on the medical history of the patients from ten Han families of central China were collected. Ophthalmic examinations including best-corrected visual acuity (BCVA), fundus photography, ultra-wide-angle sweep source optical coherence tomography (SS-OCT), and electroretinography (ERG) were performed. Adaptive optics (AO) images were acquired to evaluate the cone photoreceptor mosaic when applicable. Venous blood of the probands and their family members was collected, and DNA was subjected to sequencing based on next-generation sequencing with a custom-designed targeted gene panel PS400 for inherited retinal diseases. Validation was performed by Sanger sequencing and cosegregation. Pathogenicity was determined in accordance with the American College of Medical Genetics and Genomics (ACMG) guidelines.Results: Ten RS1 mutations, including eight missense mutations and two terminator mutations, were identified in 10 XLRS families. c.657C > A (p.C219X) was a novel mutation in this cohort. These patients showed a variety of clinical phenotypes, including fovea schisis, bullous retinoschisis, and macular or peripheral atrophy. Fifteen eyes of eight patients exhibited macular retinoschisis, and twelve eyes of seven patients exhibited peripheral retinoschisis. In addition, three patients showed asymmetrical fundus manifestations. Of importance, three patients without macular retinoschisis were misdiagnosed until genetic testing results were obtained. AO showed a decrease in cone density and loss of regularity in the cystic schisis macular of XLRS. Furthermore, the BCVA was associated with the photoreceptor inner segment and outer segment (IS/OS) thickness.Conclusion: With complicated clinical manifestations, a considerable portion of XLRS patients may present various phenotypes. It should be noted that asymmetry in fundus appearance in both eyes could lead to misdiagnosis easily. Thus, genetic testing is crucial for making a final diagnosis in those patients who are suspected of having amblyopia, bilateral or unilateral macular atrophy, or conditions presenting an asymmetric fundus appearance. In addition, the residual cone photoreceptor structure was critical for the maintenance of useful vision.
The retina is the entrance of the visual system. Although based on common biophysical principles, the dynamics of retinal neurons are quite different from their cortical counterparts, raising interesting problems for modellers. In this paper, I address some mathematically stated questions in this spirit, discussing, in particular: (1) How could lateral amacrine cell connectivity shape the spatio-temporal spike response of retinal ganglion cells? (2) How could spatio-temporal stimuli correlations and retinal network dynamics shape the spike train correlations at the output of the retina? These questions are addressed, first, introducing a mathematically tractable model of the layered retina, integrating amacrine cells’ lateral connectivity and piecewise linear rectification, allowing for computing the retinal ganglion cells receptive field together with the voltage and spike correlations of retinal ganglion cells resulting from the amacrine cells networks. Then, I review some recent results showing how the concept of spatio-temporal Gibbs distributions and linear response theory can be used to characterize the collective spike response to a spatio-temporal stimulus of a set of retinal ganglion cells, coupled via effective interactions corresponding to the amacrine cells network. On these bases, I briefly discuss several potential consequences of these results at the cortical level.
Photography, Computer applications to medicine. Medical informatics
The in situ leaf area index (LAI) measurement plays a vital role in calibrating and validating satellite LAI products. Digital hemispherical photography (DHP) is a widely used in situ forest LAI measurement method. There have been many software programs encompassing a variety of algorithms to estimate LAI from DHP. However, there is no conclusive study for an accuracy comparison among them, due to the difficulty in acquiring forest LAI reference values. In this study, we aim to use virtual (i.e., computer-simulated) broadleaf forests for the accuracy assessment of LAI algorithms in commonly used LAI software programs. Three commonly used DHP programs, including Can_Eye, CIMES, and Hemisfer, were selected since they provide estimates of both effective LAI and true LAI. Individual tree models with and without leaves were first reconstructed based on terrestrial LiDAR point clouds. Various stands were then created from these models. A ray-tracing technique was combined with the virtual forests to model synthetic DHP, for both leaf-on and leaf-off conditions. Afterward, three programs were applied to estimate PAI from leaf-on DHP and the woody area index (WAI) from leaf-off DHP. Finally, by subtracting WAI from PAI, true LAI estimates from 37 different algorithms were achieved for evaluation. The performance of these algorithms was compared with pre-defined LAI and PAI values in the virtual forests. The results demonstrated that without correcting for the vegetation clumping effect, Can_Eye, CIMES, and Hemisfer could estimate effective PAI and effective LAI consistent with each other (R<sup>2</sup> > 0.8, RMSD < 0.2). After correcting for the vegetation clumping effect, there was a large inconsistency. In general, Can_Eye more accurately estimated true LAI than CIMES and Hemisfer (with R<sup>2</sup> = 0.88 > 0.72, 0.49; RMSE = 0.45 < 0.7, 0.94; nRMSE = 15.7% < 24.21%, 32.81%). There was a systematic underestimation of PAI and LAI using Hemisfer. The most accurate algorithm for estimating LAI was identified as the P57 algorithm in Can_Eye which used the 57.5° gap fraction inversion combined with the finite-length averaging clumping correction. These results demonstrated the inconsistency of LAI estimates from DHP using different algorithms. It highlights the importance and provides a reference for standardizing the algorithm protocol for in situ forest LAI measurement using DHP.
Abstract The crowd behaviour understanding and density estimation are some of the fast‐growing fields in video surveillance. There are many techniques (detection and regression) that are used as the method of crowd analysis and estimation. In the present approach, SVR (support vector regression) is used as the basic analysis technique and the novel key‐point matching with SURF (speedup robust feature) is used as the feature extractor for moving objects in the video. The traditional linear regression methods used mainly key‐point as one of the statistical features instead of matching with consecutive frames, but we used the magnitude of the optical flow for foreground object extraction instead of inter‐frame difference. The combination of the optical flow of foreground objects and key‐point matching generates new features apart from conventional features such as areas and corners. In this new approach, key‐point pairing with linear regression is tested with the PETS2009 dataset, and performance is compared with the existing approaches.
Abstract Accurate defocus blur detection has instigated wide research interest for the last few years. However, it is still a meaningful yet challenging machine vision task, and most methods rely on prior knowledge. Convolutional neural networks have proved the huge success for different tasks within the computer vision, and machine learning flew. A simple yet effective method of defocus blur detection was proposed in this paper, which by applying the deep residual convolutional encoder‐decoder network. The aims of DRDN is to automatically generate pixel‐level predictions for defocus blur images, and reconstruct output detection results of the same size as the input, which by performing several deconvolution operations at multiple scales through the transposed convolution, and skip connection. Afterwards, we used the slide window detection strategy and traversed the input image with a certain stride. Experiments on challenging benchmarks of defocus blur detection show that our algorithm achieved state‐of‐the‐art performance, and powerfully balanced the detection accuracy, and detection time.
Rebecca G. Edwards Mayhew, Malik Y. Kahook, Leonard K. Seibold
Purpose: To remind eye care providers of the importance of obtaining vertical rasters in OCT evaluation of patients with suspected hypotony maculopathy. Observations: OCT with vertical rasters may identify chorioretinal folds that are missed by ophthalmoscopic examination, fundus photography, and traditional OCT with horizontal rasters alone. Conclusions and importance: In patients with low IOP and decreased vision, OCT imaging with horizonal and vertical rasters should be obtained to diagnose hypotony maculopathy and monitor response to treatment.
Lazaros Tsochatzidis, Lena Costaridou, Ioannis Pratikakis
Deep convolutional neural networks (CNNs) are investigated in the context of computer-aided diagnosis (CADx) of breast cancer. State-of-the-art CNNs are trained and evaluated on two mammographic datasets, consisting of ROIs depicting benign or malignant mass lesions. The performance evaluation of each examined network is addressed in two training scenarios: the first involves initializing the network with pre-trained weights, while for the second the networks are initialized in a random fashion. Extensive experimental results show the superior performance achieved in the case of fine-tuning a pretrained network compared to training from scratch.
Photography, Computer applications to medicine. Medical informatics
A. Kluge and G. Richter are meeting here around the issue of chance: as time passes, it irreversibly precipitates certain events among a series of possibilities that could have been realized. The unpredictable and brutal nature of events raises following questions: Why does the world take one turn rather than another? How do we resist the fatal flow of time? While mastering time is an illusory endeavor, art allows us to reintroduce possibilities that history has removed. Artists invite us through their work to re-think history – both its unflinching certainties and its alternatives. They thus subdue chance, reintroduce calm into the present and allow us to postpone direct analyses to durably transform our representations.
With the panoramic video installation ‘Decidophobia’, I endeavour to point to the difficulty of making decisions and the confrontation with overwhelming choices in our satiated society of today with the means of media arts I aim to raise the awareness for the great gift we have: The freedom to choose in many situations of life; but as well being overwhelmed and hindered by an extensive choice and the difficult task of making the right decision. This finally might raise counterfactual thoughts, the retrospective considerations of what would be today, if we decided differently at a turning point at an earlier stage of our life.
In this work, the viewer is confronted with the difficulty of choice. A perfect labyrinth is all around, no orientation is possible, paths appear and disappear again, vanish completely, but new possibilities open up. The soundtrack underlines the visual experience: passers-by are expressing their confusion by questioning where they actually are, where they should go to, which path to take and where they came from. A Babel-like chatter in nine different languages from different directions enhances the confusion. It is difficult to focus on one voice, on a familiar language, which leaves the viewer with the impression of a missed conversation, a missed opportunity.