Uncertainty quantification via localized gradients for deep learning-based medical image assessments.

Objective.Deep learning models that aid in medical image assessment tasks must be both accurate and reliable to be deployed within clinical settings. While deep learning models have been shown to be highly accurate across a variety of tasks, measures that indicate the reliability of these models are less established. Increasingly, uncertainty quantification (UQ) methods are being introduced to inform users on the reliability of model outputs. However, most existing methods cannot be augmented to previously validated models because they are not post hoc, and they change a model's output. In this work, we overcome these limitations by introducing a novel post hoc UQ method, termedLocal Gradients UQ, and demonstrate its utility for deep learning-based metastatic disease delineation.Approach.This method leverages a trained model's localized gradient space to assess sensitivities to trained model parameters. We compared the Local Gradients UQ method to non-gradient measures defined using model probability outputs. The performance of each uncertainty measure was assessed in four clinically relevant experiments: (1) response to artificially degraded image quality, (2) comparison between matched high- and low-quality clinical images, (3) false positive (FP) filtering, and (4) correspondence with physician-rated disease likelihood.Main results.(1) Response to artificially degraded image quality was enhanced by the Local Gradients UQ method, where the median percent difference between matching lesions in non-degraded and most degraded images was consistently higher for the Local Gradients uncertainty measure than the non-gradient uncertainty measures (e.g. 62.35% vs. 2.16% for additive Gaussian noise). (2) The Local Gradients UQ measure responded better to high- and low-quality clinical images (p< 0.05 vsp> 0.1 for both non-gradient uncertainty measures). (3) FP filtering performance was enhanced by the Local Gradients UQ method when compared to the non-gradient methods, increasing the area under the receiver operating characteristic curve (ROC AUC) by 20.1% and decreasing the false positive rate by 26%. (4) The Local Gradients UQ method also showed more favorable correspondence with physician-rated likelihood for malignant lesions by increasing ROC AUC for correspondence with physician-rated disease likelihood by 16.2%.Significance. In summary, this work introduces and validates a novel gradient-based UQ method for deep learning-based medical image assessments to enhance user trust when using deployed clinical models.

Breast cancer risk assessment and risk distribution in 3,491 Slovenian women invited for screening at the age of 50; a population-based cross-sectional study.

BACKGROUND: The evidence shows that risk-based strategy could be implemented to avoid unnecessary harm in mammography screening for breast cancer (BC) using age-only criterium. Our study aimed at identifying the uptake of Slovenian women to the BC risk assessment invitation and assessing the number of screening mammographies in case of risk-based screening. PATIENTS AND METHODS: A cross-sectional population-based study enrolled 11,898 women at the age of 50, invited to BC screening. The data on BC risk factors, including breast density from the first 3,491 study responders was collected and BC risk was assessed using the Tyrer-Cuzick algorithm (version 8) to classify women into risk groups (low, population, moderately increased, and high risk group). The number of screening mammographies according to risk stratification was simulated. RESULTS: 57% (6,785) of women returned BC risk questionnaires. When stratifying 3,491 women into risk groups, 34.0% were assessed with low, 62.2% with population, 3.4% with moderately increased, and 0.4% with high 10-year BC risk. In the case of potential personalised screening, the number of screening mammographies would drop by 38.6% compared to the current screening policy. CONCLUSIONS: The study uptake showed the feasibility of risk assessment when inviting women to regular BC screening. 3.8% of Slovenian women were recognised with higher than population 10-year BC risk. According to Slovenian BC guidelines they may be screened more often. Overall, personalised screening would decrease the number of screening mammographies in Slovenia. This information is to be considered when planning the pilot and assessing the feasibility of implementing population risk-based screening.

Uncertainty estimation for deep learning-based pectoral muscle segmentation via Monte Carlo dropout.

Objective. Deep Learning models are often susceptible to failures after deployment. Knowing when your model is producing inadequate predictions is crucial. In this work, we investigate the utility of Monte Carlo (MC) dropout and the efficacy of the proposed uncertainty metric (UM) for flagging of unacceptable pectoral muscle segmentations in mammograms.Approach. Segmentation of pectoral muscle was performed with modified ResNet18 convolutional neural network. MC dropout layers were kept unlocked at inference time. For each mammogram, 50 pectoral muscle segmentations were generated. The mean was used to produce the final segmentation and the standard deviation was applied for the estimation of uncertainty. From each pectoral muscle uncertainty map, the overall UM was calculated. To validate the UM, a correlation between the dice similarity coefficient (DSC) and UM was used. The UM was first validated in a training set (200 mammograms) and finally tested in an independent dataset (300 mammograms). ROC-AUC analysis was performed to test the discriminatory power of the proposed UM for flagging unacceptable segmentations.Main results. The introduction of dropout layers in the model improved segmentation performance (DSC = 0.95 ± 0.07 versus DSC = 0.93 ± 0.10). Strong anti-correlation (r= -0.76,p< 0.001) between the proposed UM and DSC was observed. A high AUC of 0.98 (97% specificity at 100% sensitivity) was obtained for the discrimination of unacceptable segmentations. Qualitative inspection by the radiologist revealed that images with high UM are difficult to segment.Significance. The use of MC dropout at inference time in combination with the proposed UM enables flagging of unacceptable pectoral muscle segmentations from mammograms with excellent discriminatory power.

Measurements of hair temperature avalanche effect with alexandrite and Nd:YAG hair removal lasers.

BACKGROUND AND OBJECTIVES: In this study, we investigate the photothermal response of human hair using a pulsed laser source employed in the hair removal treatment. The purpose is to understand the dynamics behind the most common clinical practice to better define the salient features that may contribute to the efficiency of the process. STUDY DESIGN/MATERIALS AND METHODS: Temperature changes of hair samples (dark brown color) from a human scalp (skin type Fitpatrick II) were measured by a thermal camera following irradiation with single and multiple neodymium: yttrium-aluminum-garnet (Nd:YAG) (1064 nm) and alexandrite (755 nm) laser pulses. Particularly, the hair was treated with an individual laser pulse of a sufficiently high fluence, or with a series of lower fluence laser pulses. We investigated the temperature increase in a broad range of fluence and number of pulses. From the data analysis we extrapolated important parameters such as thermal gain and threshold fluence that can be used for determining optimal parameters for the hair removal procedure. Our experimental investigations and hypothesis were supported by a numerical simulation of the light-matter interaction in a skin-hair model, and by optical transmittance measurements of the irradiated hair. RESULTS: An enhancement of the temperature response of the irradiated hair, that deviates from the linear behavior, is observed when hair is subjected to an individual laser pulse of a sufficiently high fluence or to a series of lower fluence laser pulses. Here, we defined the nonlinear and rapid temperature built-up as an avalanche effect. We estimated the threshold fluence at which this process takes place to be at 10 and 2.5 J/cm2 for 1064 and 755 nm laser wavelengths, respectively. The thermal gain expressed by the degree of the deviation from the linear behavior can be higher than 2 when low laser fluence and multiple laser pulses are applied (n = 50). The comparison of the calculated gain for the two different laser wavelengths and the number of pulses reveals a much higher efficiency when low fluence and multiple pulses are delivered. The avalanche effect manifests when the hair temperature exceeds 45°C. The enhanced temperature increase during the subsequent delivery of laser pulses could be ascribed to the temperature-induced changes in the hair's structural properties. Simulations of the hair temperature under Nd:YAG and alexandrite irradiation indicate that the avalanche phenomenon observed in the hair suspended in air may apply also to the hair located within the skin matrix. Namely, for the same fluence, similar temperature increase was obtained also for the hair located within the skin. CONCLUSION: The observed "avalanche" effect may contribute to the reported clinical efficacy of laser hair removal and may at least partially explain the observed efficacy of the brushing hair removal procedures where laser fluence is usually low. The repeated irradiation during the brushing procedure may lead to an avalanche-like gradual increase of the hair's thermal response resulting in sufficiently high final hair temperatures as required for effective hair reduction.

Quantitative imaging biomarkers of immune-related adverse events in immune-checkpoint blockade-treated metastatic melanoma patients: a pilot study.

PURPOSE: To develop quantitative molecular imaging biomarkers of immune-related adverse event (irAE) development in malignant melanoma (MM) patients receiving immune-checkpoint inhibitors (ICI) imaged with 18F-FDG PET/CT. METHODS: 18F-FDG PET/CT images of 58 MM patients treated with anti-PD-1 or anti-CTLA-4 ICI were retrospectively analyzed for indication of irAE. Three target organs, most commonly affected by irAE, were considered: bowel, lung, and thyroid. Patient charts were reviewed to identify which patients experienced irAE, irAE grade, and time to irAE diagnosis. Target organs were segmented using a convolutional neural network (CNN), and novel quantitative imaging biomarkers - SUV percentiles (SUVX%) of 18F-FDG uptake within the target organs - were correlated with the clinical irAE status. Area under the receiver-operating characteristic curve (AUROC) was used to quantify irAE detection performance. Patients who did not experience irAE were used to establish normal ranges for target organ 18F-FDG uptake. RESULTS: A total of 31% (18/58) patients experienced irAE in the three target organs: bowel (n=6), lung (n=5), and thyroid (n=9). Optimal percentiles for identifying irAE were bowel (SUV95%, AUROC=0.79), lung (SUV95%, AUROC=0.98), and thyroid (SUV75%, AUROC=0.88). Optimal cut-offs for irAE detection were bowel (SUV95%>2.7 g/mL), lung (SUV95%>1.7 g/mL), and thyroid (SUV75%>2.1 g/mL). Normal ranges (95% confidence interval) for the SUV percentiles in patients without irAE were bowel [1.74, 2.86 g/mL], lung [0.73, 1.46 g/mL], and thyroid [0.86, 1.99 g/mL]. CONCLUSIONS: Increased 18F-FDG uptake within irAE-affected organs provides predictive information about the development of irAE in MM patients receiving ICI and represents a potential quantitative imaging biomarker for irAE. Some irAE can be detected on 18F-FDG PET/CT well before clinical symptoms appear.