Development of a high-throughput screening platform to identify new therapeutic agents for Medulloblastoma Group 3

Pediatric brain tumors (PBTs) represent about 25 % of all pediatric cancers and are the most common solid tumors in children and adolescents. Medulloblastoma (MB) is the most frequently occurring malignant PBT, accounting for almost 10 % of all pediatric cancer deaths. MB Group 3 (MB G3) accounts for 25-30 % of all MB cases and has the worst outcome, particularly when associated with MYC amplification. However, no targeted treatments for this group have been developed so far. Here we describe a unique high throughput screening (HTS) platform specifically designed to identify new therapies for MB G3. The platform incorporates optimized and validated 2D and 3D efficacy and toxicity models, that account for tumor heterogenicity, limited efficacy and unacceptable toxicity from the very early stage of drug discovery. The platform has been validated by conducting a pilot HTS campaign with a 1280 lead-like compound library. Results showed 8 active compounds, targeting MB reported targets and several are currently approved or in clinical trials for pediatric patients with PBTs, including MB. Moreover, hits were combined to avoid tumor resistance, identifying 3 synergistic pairs, one of which is currently under clinical study for recurrent MB and other PBTs.

Using deep learning for dermatologist-level detection of suspicious pigmented skin lesions from wide-field images

A reported 96,480 people were diagnosed with melanoma in the United States in 2019, leading to 7230 reported deaths. Early-stage identification of suspicious pigmented lesions (SPLs) in primary care settings can lead to improved melanoma prognosis and a possible 20-fold reduction in treatment cost. Despite this clinical and economic value, efficient tools for SPL detection are mostly absent. To bridge this gap, we developed an SPL analysis system for wide-field images using deep convolutional neural networks (DCNNs) and applied it to a 38,283 dermatological dataset collected from 133 patients and publicly available images. These images were obtained from a variety of consumer-grade cameras (15,244 nondermoscopy) and classified by three board-certified dermatologists. Our system achieved more than 90.3% sensitivity (95% confidence interval, 90 to 90.6) and 89.9% specificity (89.6 to 90.2%) in distinguishing SPLs from nonsuspicious lesions, skin, and complex backgrounds, avoiding the need for cumbersome individual lesion imaging. We also present a new method to extract intrapatient lesion saliency (ugly duckling criteria) on the basis of DCNN features from detected lesions. This saliency ranking was validated against three board-certified dermatologists using a set of 135 individual wide-field images from 68 dermatological patients not included in the DCNN training set, exhibiting 82.96% (67.88 to 88.26%) agreement with at least one of the top three lesions in the dermatological consensus ranking. This method could allow for rapid and accurate assessments of pigmented lesion suspiciousness within a primary care visit and could enable improved patient triaging, utilization of resources, and earlier treatment of melanoma.

Computer-aided classification of suspicious pigmented lesions using wide-field images

BACKGROUND AND OBJECTIVE

Early identification of melanoma is conducted through whole-body visual examinations to detect suspicious pigmented lesions, a situation that fluctuates in accuracy depending on the experience and time of the examiner. Computer-aided diagnosis tools for skin lesions are typically trained using pre-selected single-lesion images, taken under controlled conditions, which limits their use in wide-field scenes. Here, we propose a computer-aided classifier system with such input conditions to aid in the rapid identification of suspicious pigmented lesions at the primary care level.

METHODS

133 patients with a multitude of skin lesions were recruited for this study. All lesions were examined by a board-certified dermatologist and classified into "suspicious" and "non-suspicious". A new clinical database was acquired and created by taking Wide-Field images of all major body parts with a consumer-grade camera under natural illumination condition and with a consistent source of image variability. 3-8 images were acquired per patient on different sites of the body, and a total of 1759 pigmented lesions were extracted. A machine learning classifier was optimized and build into a computer aided classification system to binary classify each lesion using a suspiciousness score.

RESULTS

In a testing set, our computer-aided classification system achieved a sensitivity of 100% for suspicious pigmented lesions that were later confirmed by dermoscopy examination ("SPL_A") and 83.2% for suspicious pigmented lesions that were not confirmed after examination ("SPL_B"). Sensitivity for non-suspicious lesions was 72.1%, and accuracy was 75.9%. With these results we defined a suspiciousness score that is aligned with common macro-screening (naked eye) practices.

CONCLUSIONS

This work demonstrates that wide-field photography combined with computer-aided classification systems can distinguish suspicious from non-suspicious pigmented lesions, and might be effective to assess the severity of a suspicious pigmented lesions. We believe this approach could be useful to support skin screenings at a population-level.

Epileptogenic Zone Localization With 18FDG PET Using a New Dynamic Parametric Analysis

Introduction: [¹⁸F]fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography (PET) is part of the regular preoperative work-up in medically refractory epilepsy. As a complement to visual evaluation of PET, statistical parametric maps can help in the detection of the epileptogenic zone (EZ). However, software packages currently available are time-consuming and little intuitive for physicians. We develop a user-friendly software (referred as PET-analysis) for EZ localization in PET studies that allows dynamic real-time statistical parametric analysis. To evaluate its performance, the outcome of PET-analysis was compared with the results obtained by visual assessment and Statistical Parametric Mapping (SPM). Methods: Thirty patients with medically refractory epilepsy who underwent presurgical ¹⁸F-FDG PET with good post-operative outcomes were included. The ¹⁸F-FDG PET studies were evaluated by visual assessment, with SPM8 and PET-analysis. In SPM, parametric T-maps were thresholded at corrected p < 0.05 and cluster size k = 50 and at uncorrected p < 0.001 and k = 100 (the most used parameters in the literature). Since PET-analysis rapidly processes different threshold combinations, T-maps were thresholded with multiple p-value and different clusters sizes. The presurgical EZ identified by visual assessment, SPM and PET-analysis was compared to the confirmed EZ according to post-surgical follow-up. Results: PET-analysis obtained 66.7% (20/30) of correctly localizing studies, comparable to the 70.0% (21/30) achieved by visual assessment and significantly higher (p < 0.05) than that obtained with the SPM threshold p < 0.001/k = 100, of 36.7% (11/30). Only one study was positive, albeit non-localizing, with the SPM threshold corrected p < 0.05/k = 50. Concordance was substantial for PET-analysis (κ = 0.643) and visual interpretation (κ = 0.622), being fair for SPM (κ = 0.242). Conclusion: Compared to SPM with the fixed standard parameters, PET-analysis may be superior in EZ localization with its easy and rapid processing of different threshold combinations. The results of this initial proof-of-concept study validate the clinical use of PET-analysis as a robust objective complementary tool to visual assessment for EZ localization.

Diffeomorphic Lung Registration Using Deep CNNs and Reinforced Learning

Image registration is a well-known problem in the field of medical imaging. In this paper, we focus on the registration of chest inspiratory and expiratory computed tomography (CT) scans from the same patient. Our method recovers the diffeomorphic elastic displacement vector field (DVF) by jointly regressing the direct and the inverse transformation. Our architecture is based on the RegNet network but we implement a reinforced learning strategy that can accommodate a large training dataset. Our results show that our method performs with a lower estimation error for the same number of epochs than the RegNet approach.

PISCOM: a new procedure for epilepsy combining ictal SPECT and interictal PET

PURPOSE

We present a modified version of the SISCOM procedure that uses interictal PET instead of interictal SPECT for seizure onset zone localization. We called this new nuclear imaging processing technique PISCOM (PET interictal subtracted ictal SPECT coregistered with MRI).

METHODS

We retrospectively studied 23 patients (age range 4-61 years) with medically refractory epilepsy who had undergone MRI, ictal SPECT, interictal SPECT and interictal FDG PET and who had been seizure-free for at least 2 years after surgical treatment. FDG PET images were reprocessed (rFDG PET) to assimilate SPECT features for image subtraction. Interictal SPECT and rFDG PET were compared using statistical parametric mapping (SPM). PISCOM and SISCOM images were evaluated visually and using an automated volume of interest-based analysis. The results of the two studies were compared with each other and with the known surgical resection site.

RESULTS

SPM showed no significant differences in cortical activity between SPECT and rFDG PET images. PISCOM and SISCOM showed equivalent results in 17 of 23 patients (74%). The seizure onset zone was successfully identified in 19 patients (83%) by PISCOM and in 17 (74%) by SISCOM: in 15 patients (65%) the two techniques showed concordant successful results. The volume of interest-based analysis showed no significant differences between PISCOM and SISCOM in identifying the extension of the seizure onset zone. However, PISCOM showed a lower amount of indeterminate activity due to propagation, background or artefacts.

CONCLUSION

Preliminary findings of this initial proof-of-concept study suggest that perfusion and glucose metabolism in the cerebral cortex can be correlated and that PISCOM may be a valid technique for identification of the seizure onset zone. However, further studies are needed to validate these results.

Quantification of Very Low Concentrations of Leukocyte Suspensions In Vitro by High-Frequency Ultrasound

Accurate measurement of very low cerebrospinal fluid (CSF) white blood cell (WBC) concentration is key to the diagnosis of bacterial meningitis, lethal if not promptly treated. Here we show that high frequency ultrasound (HFUS) can detect CSF WBC in vitro in concentrations relevant to meningitis diagnosis with a much finer precision than gold standard manual counting in a Fuchs-Rosenthal chamber. WBC concentrations in a mock CSF model, in the range 0-50 WBC/μL, have been tested and compared to gold standard ground truth. In this range, excellent agreement (Cohen's kappa [κ] = 0.78-90) (Cohen 1960) was observed between HFUS and the gold standard method. The presented experimental set-up allowed us to detect WBC concentrations as low as 2 cells/μL. HFUS shows promise as a low-cost, reliable and automated technology to measure very low CSF WBC concentrations for the diagnosis of early meningitis.

Including anatomical and functional information in MC simulation of PET and SPECT brain studies. Brain-VISET: a voxel-based iterative method

Monte Carlo (MC) simulation provides a flexible and robust framework to efficiently evaluate and optimize image processing methods in emission tomography. In this work we present Brain-VISET (Voxel-based Iterative Simulation for Emission Tomography), a method that aims to simulate realistic [ (99m) Tc]-SPECT and [ (18) F]-PET brain databases by including anatomical and functional information. To this end, activity and attenuation maps generated using high-resolution anatomical images from patients were used as input maps in a MC projector to simulate SPECT or PET sinograms. The reconstructed images were compared with the corresponding real SPECT or PET studies in an iterative process where the activity inputs maps were being modified at each iteration. Datasets of 30 refractory epileptic patients were used to assess the new method. Each set consisted of structural images (MRI and CT) and functional studies (SPECT and PET), thereby allowing the inclusion of anatomical and functional variability in the simulation input models. SPECT and PET sinograms were obtained using the SimSET package and were reconstructed with the same protocols as those employed for the clinical studies. The convergence of Brain-VISET was evaluated by studying the behavior throughout iterations of the correlation coefficient, the quotient image histogram and a ROI analysis comparing simulated with real studies. The realism of generated maps was also evaluated. Our findings show that Brain-VISET is able to generate realistic SPECT and PET studies and that four iterations is a suitable number of iterations to guarantee a good agreement between simulated and real studies.