This study evaluated whether deep learning frameworks trained in large datasets can help non-dermatologist physicians improve their accuracy in categorizing the seven most common pigmented skin lesions. Open-source skin images were downloaded from the International Skin Imaging Collaboration (ISIC) archive. Different deep neural networks (DNNs) (n = 8) were trained based on a random dataset constituted of 8015 images. A test set of 2003 images was used to assess the classifiers’ performance at low (300 × 224 RGB) and high (600 × 450 RGB) image resolution and aggregated data (age, sex and lesion localization). We also organized two different contests to compare the DNN performance to that of general practitioners by means of unassisted image observation. Both at low and high image resolution, the DNN framework differentiated dermatological images with appreciable performance. In all cases, the accuracy was improved when adding clinical data to the framework. Finally, the least accurate DNN outperformed general practitioners. The physician’s accuracy was statistically improved when allowed to use the output of this algorithmic framework as guidance. DNNs are proven to be high performers as skin lesion classifiers and can improve general practitioner diagnosis accuracy in a routine clinical scenario.
Artificial intelligence can be a key tool in the context of assisting in the diagnosis of dermatological conditions, particularly when performed by general practitioners with limited or no access to high resolution optical equipment. This study evaluates the performance of deep convolutional neural networks (DNNs) in the classification of seven pigmented skin lesions. Additionally, it assesses the improvement ratio in the classification performance when utilized by general practitioners. Open-source skin images were downloaded from the ISIC archive. Different DNNs (n=8) were trained based on a random dataset constituted by 8,015 images. A test set of 2,003 images has been used to assess the classifiers performance at low (300 x 224 RGB) and high (600 x 450 RGB) image resolution and aggregated clinical data (age, sex and lesion localization). We have also organized two different contests to compare the DNNs performance to that of general practitioners by means of unassisted image observation. Both at low and high image resolution, the DNNs framework being trained differentiated dermatological images with appreciable performance. In all cases, accuracy has been improved when adding clinical data to the framework. Finally, the lowest accurate DNN outperformed general practitioners. Physicians accuracy was statistically improved when allowed to use the output of this algorithmic framework as guidance. DNNS are proven to be high performers as skin lesion classifiers. The aim is to include these AI tools in the context of general practitioners whilst improving their diagnosis accuracy in a routine clinical scenario when or where the use of high-resolution equipment is not accessible.
1548 Background: Synthetic fingerprints integrate clinical data within computational models allowing the identification of particular clinical subpopulations at a given moment. We here describe a deep learning strategy to detect super-responder and super-survivor patients with squamous NSCLC by setting up synthetic fingerprints and using unsupervised deep learning frameworks (UDLF). Methods: Through www.projectdatasphere.org, we accessed the control arm clinical data (N = 548) of the randomised phase III SQUIRE trial (NCT00981058). This trial included patients with stage IV squamous NSCLC who had not received previous chemotherapy. These patients were treated with gemcitabine 1,250 mg/m2 (IV, 30-min infusion, d1/d8) and cisplatin 75 mg/m2 (IV, 120 min infusion, d1) on a 3-week cycle for a maximum of six cycles. Synthetic fingerprints resulted of the integration of 180 features collected during the first 3 cycles including demographics, medical history, physical exam, concomitant medication, histopathology, PK parameters, adverse events and common labs. These fingerprints were used as input for the UDLF. The resultant clusters were correlated with overall-response rate (ORR) and overall survival (OS). Results: After missing data removal and feature standardization, 192 patients were eligible for the study. The UDLF was able to generate two different clusters: P0 (n = 107) and P1 (n = 84). ORR was higher in the P1 than in the P0 cluster (mean 41.6% [95% CI 31.7-52.3] vs. 28.0% [95% CI 20.4-37.2]; p = 0.04). OS was significantly longer in the P1 than in the P0 cluster (median 13.2 months vs. 9.7 months; hazard ratio 1.56 [95% CI 1.12-2.17; p = 0.008]). Feature contribution analysis showed that P1 had more patients and more events of grade III/IV neutropenia. In contrast, P0 had more patients and more events of grade III/IV nausea and vomiting. Other major differences were observed on vital signs (SBP, DBP, HR, RR, Temp), concomitant medication (osmotically-active laxatives, dexamethasone, furosemide, granisetron and ondansetron) and in hematological (RBC, HGB, HCT, MCV, WBC, neutrophils, monocytes, lymphocytes) and biochemistry (albumin, globulins, ALP, LDH, creatinine, BUN, urea, sodium, magnesium and phosphate) tests. Conclusions: Our findings show that synthetic fingerprints and subsequent deep learning analysis can be of use to identify patients with clinical characteristics associated with high-response rate and long-term survival.
1549 Background: Predicting the clinical course of metastatic disease remains a key challenge in CRC. Estimating prognosis of these late-stage patients can avoid undertreatment or overtreatment and also guide the follow-up intensity. This study has investigated the ability of an artificial intelligence-based analytical tool to identify those mCRC patients with high risk of disease progression and mortality based on their clinical parameters. Methods: Through www.projectdatasphere.org we accessed datasets of two randomised phase III trials including chemo-naïve (NCT00364013, n = 1183 patients) and chemo-refractory (NCT00113763, N = 483) mCRC patients. We generated synthetic fingerprints (SF) for each patient through the integration of 44 clinical features (demographics, anthropometrics, medical history, blood tests and treatment characteristics) collected, respectively, during the screening phase and the first month of inclusion in each trial. These SF were then input into a deep learning framework (DLF) to identify subgroup of patients based on their similarities. The resultant clusters were correlated with progression-free survival (PFS) and overall survival (OS). Results: After discarding missing data, 861 chemo-naïve and 341 chemo-resistant mCRC patients were eligible for the study. In the chemo-naïve cohort, the SF/DLF system was able to detect two different clusters: C0 (n = 31) and C1 (n = 830). Patients in C0 had a higher risk of progression (median PFS 6.2 months vs. 9.1 months; hazard ratio 1.83, 95% CI 1.16-2.88; p = 0.008) and death (median OS 13.2 months vs. 20.1 months; hazard ratio 2.84, 95% CI 1.68-4.80; p < 0.001) compared to patients in C1. When applied to the chemo-resistant cohort, the SF/DLF system was again able to identify two different clusters: P0 (n = 159) and P1 (n = 182). Patients in P0 had a higher risk of progression (median PFS 1.7 months vs. 1.8 months; hazard ratio 1.32, 95% CI 1.05-1.67; p < 0.001) and death (median OS 6.1 months vs. 6.8 months; hazard ratio 1.34, 95% CI 1.07-1.68; p = 0.01) compared to patients in P1. In both cases, feature contribution analysis showed that major differences between clusters were related to clinical status, anthropometrics and haematological and biochemistry tests. Conclusions: Our SF/DLF system can identify mCRC subtypes based on distinct clinical features that correlate with higher risk of progression and death. Further work is required to validate this approach as a novel prognostic biomarker tool for monitoring mCRC patients.
Trial design: Adults with radiologic, histologic, or cytologic confirmed HCC, Child-Pugh class A, ECOG performance status 0 or 1, Barcelona Clinic Liver Cancer stage C or stage B not amenable to locoregional therapy or refractory to locoregional therapy and not amenable to curative treatment, with !1 measurable lesion per RECIST v1.1 confirmed by blinded independent central review (BICR), and adequate organ function are eligible. Patients previously treated with locoregional therapy ( 4 weeks of the first dose of study treatment for HCC), systemic chemotherapy for HCC, or an antiePD-1/PD-L1/PD-L2 agent or another stimulatory or coinhibitory T-cell receptore directed agent are ineligible. The study is divided into 2 phases: The first phase, the Safety Lead-in Phase, will evaluate the tolerability of the predefined recommended phase 2 dose of MK-1308A plus lenvatinib. The second phase, the Efficacy Expansion Phase, will evaluate the safety and efficacy of MK-1308A plus lenvatinib. The Safety Lead-in Phase will include 6-20 patients, depending on the occurrence of dose limiting toxicities (DLTs) and will be considered complete when w10 patients eligible for DLT evaluation are allocated to and treated at dose level 0 or À1. Dose level 0 consists of MK-1308A (quavonlimab 25 mg/pembrolizumab 400 mg IV every 6 weeks [Q6W]) plus lenvatinib (8 mg [body weight (BW) < 60 kg] or 12 mg [BW !60 kg] orally once daily [QD]). Dose level À1 consists of MK-1308A (quavonlimab 25 mg/pembrolizumab 400 mg IV Q6W) plus lenvatinib (4 mg BW < 60 kg or 8 mg BW !60 kg orally QD). If dose level 0 is tolerable, the Efficacy Expansion Phase will be opened, and w100-104 patients could be enrolled (110 patients total at dose level 0 in the Safety Lead-in and Efficacy Expansion Phases combined). In the Efficacy Expansion Phase, the primary end points are safety and overall response rate (ORR) per RECIST v1.1 by BICR. Secondary end points are duration of response (DOR), disease control rate (DCR), progression-free survival (PFS), time to progression (TTP), all assessed per RECIST v1.1 by BICR; overall survival; and ORR, DOR, DCR, PFS, and TTP, all assessed per modified RECIST by BICR. Recruitment for this study began in March 2021.Clinical trial identification: NCT04740307.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.