Publications

2023
Guihong Wan, Bonnie Leung, Nga Nguyen, Mia S DeSimone, Feng Liu, Min Seok Choi, Diane Ho, Valerie Laucks, Stacey Duey, Ryan J Sullivan, Genevieve M Boland, Nicole R LeBoeuf, David Liu, Alexander Gusev, Shawn G Kwatra, Peter K Sorger, Kun-Hsing Yu, and Yevgeniy R Semenov. 2023. “The impact of stage-related features in melanoma recurrence prediction: A machine learning approach.” JAAD Int, 10, Pp. 28-30.
2022
Evan Brociner, Kun-Hsing Yu, Isaac S Kohane, and McGreggor Crowley. 2022. “Association of Race and Socioeconomic Disadvantage With Missed Telemedicine Visits for Pediatric Patients During the COVID-19 Pandemic.” JAMA Pediatr, 176, 9, Pp. 933-935.Abstract
This comparative effectiveness study examines whether race and ethnicity and socioeconomic disadvantage are factors associated with missing telemedicine visits during the COVID-19 pandemic among pediatric patients in Massachusetts.
Chenyue Lu, Di Jin, Nathan Palmer, Kathe Fox, Isaac S Kohane, Jordan W Smoller, and Kun-Hsing Yu. 2022. “Large-scale real-world data analysis identifies comorbidity patterns in schizophrenia.” Transl Psychiatry, 12, 1, Pp. 154.Abstract
Schizophrenia affects >3.2 million people in the USA. However, its comorbidity patterns have not been systematically characterized in real-world populations. To address this gap, we conducted an observational study using a cohort of 86 million patients in a nationwide health insurance dataset. We identified participants with schizophrenia and those without schizophrenia matched by age, sex, and the first three digits of zip code. For each phenotype encoded in phecodes, we compared their prevalence in schizophrenia patients and the matched non-schizophrenic participants, and we performed subgroup analyses stratified by age and sex. Results show that anxiety, posttraumatic stress disorder, and substance abuse commonly occur in adolescents and young adults prior to schizophrenia diagnoses. Patients aged 60 and above are at higher risks of developing delirium, alcoholism, dementia, pelvic fracture, and osteomyelitis than their matched controls. Type 2 diabetes, sleep apnea, and eating disorders were more prevalent in women prior to schizophrenia diagnosis, whereas acute renal failure, rhabdomyolysis, and developmental delays were found at higher rates in men. Anxiety and obesity are more commonly seen in patients with schizoaffective disorders compared to patients with other types of schizophrenia. Leveraging a large-scale insurance claims dataset, this study identified less-known comorbidity patterns of schizophrenia and confirmed known ones. These comorbidity profiles can guide clinicians and researchers to take heed of early signs of co-occurring diseases.
Oren Miron and Kun-Hsing Yu. 2022. “Outdoor mass gathering events and SARS-CoV-2 infection in Catalonia (North-East Spain).” Lancet Reg Health Eur, 15, Pp. 100350.
Guihong Wan, Nga Nguyen, Feng Liu, Mia S DeSimone, Bonnie W Leung, Ahmad Rajeh, Michael R Collier, Min Seok Choi, Munachimso Amadife, Kimberly Tang, Shijia Zhang, Jordan S Phillipps, Ruple Jairath, Nora A Alexander, Yining Hua, Meng Jiao, Wenxin Chen, Diane Ho, Stacey Duey, István Balázs Németh, Gyorgy Marko-Varga, Jeovanis Gil Valdés, David Liu, Genevieve M Boland, Alexander Gusev, Peter K Sorger, Kun-Hsing Yu, and Yevgeniy R Semenov. 2022. “Prediction of early-stage melanoma recurrence using clinical and histopathologic features.” NPJ Precis Oncol, 6, 1, Pp. 79.Abstract
Prognostic analysis for early-stage (stage I/II) melanomas is of paramount importance for customized surveillance and treatment plans. Since immune checkpoint inhibitors have recently been approved for stage IIB and IIC melanomas, prognostic tools to identify patients at high risk of recurrence have become even more critical. This study aims to assess the effectiveness of machine-learning algorithms in predicting melanoma recurrence using clinical and histopathologic features from Electronic Health Records (EHRs). We collected 1720 early-stage melanomas: 1172 from the Mass General Brigham healthcare system (MGB) and 548 from the Dana-Farber Cancer Institute (DFCI). We extracted 36 clinicopathologic features and used them to predict the recurrence risk with supervised machine-learning algorithms. Models were evaluated internally and externally: (1) five-fold cross-validation of the MGB cohort; (2) the MGB cohort for training and the DFCI cohort for testing independently. In the internal and external validations, respectively, we achieved a recurrence classification performance of AUC: 0.845 and 0.812, and a time-to-event prediction performance of time-dependent AUC: 0.853 and 0.820. Breslow tumor thickness and mitotic rate were identified as the most predictive features. Our results suggest that machine-learning algorithms can extract predictive signals from clinicopathologic features for early-stage melanoma recurrence prediction, which will enable the identification of patients that may benefit from adjuvant immunotherapy.
Tomotaka Ugai, Li Liu, Fred K Tabung, Tsuyoshi Hamada, Benjamin W Langworthy, Naohiko Akimoto, Koichiro Haruki, Yasutoshi Takashima, Kazuo Okadome, Hidetaka Kawamura, Melissa Zhao, Seyed Mostafa Mousavi Kahaki, Jonathan N Glickman, Jochen K Lennerz, Xuehong Zhang, Andrew T Chan, Charles S Fuchs, Mingyang Song, Molin Wang, Kun-Hsing Yu, Marios Giannakis, Jonathan A Nowak, Jeffrey A Meyerhardt, Kana Wu, Shuji Ogino, and Edward L Giovannucci. 2022. “Prognostic role of inflammatory diets in colorectal cancer overall and in strata of tumor-infiltrating lymphocyte levels.” Clin Transl Med, 12, 11, Pp. e1114.Abstract
BACKGROUND: Certain dietary patterns can elicit systemic and intestinal inflammatory responses, which may influence adaptive anti-tumor immune responses and tumor behavior. We hypothesized that pro-inflammatory diets might be associated with higher colorectal cancer mortality and that the association might be stronger for tumors with lower immune responses. METHODS: We calculated an empirical dietary inflammatory pattern (EDIP) score in 2829 patients among 3988 incident rectal and colon carcinoma cases in the Nurses' Health Study and Health Professionals Follow-up Study. Using Cox proportional hazards regression analyses, we examined the prognostic association of EDIP scores and whether it might be modified by histopathologic immune reaction (in 1192 patients with available data). RESULTS: Higher EDIP scores after colorectal cancer diagnosis were associated with worse survival, with multivariable-adjusted hazard ratios (HRs) for the highest versus lowest tertile of 1.41 (95% confidence interval [CI]: 1.13-1.77; Ptrend = 0.003) for 5-year colorectal cancer-specific mortality and 1.44 (95% CI, 1.19-1.74; Ptrend = 0.0004) for 5-year all-cause mortality. The association of post-diagnosis EDIP scores with 5-year colorectal cancer-specific mortality differed by degrees of tumor-infiltrating lymphocytes (TIL; Pinteraction = .002) but not by three other lymphocytic reaction patterns. The multivariable-adjusted, 5-year colorectal cancer-specific mortality HRs for the highest versus lowest EDIP tertile were 1.59 (95% CI: 1.01-2.53) in TIL-absent/low cases and 0.48 (95% CI: 0.16-1.48) in TIL-intermediate/high cases. CONCLUSIONS: Pro-inflammatory diets after colorectal cancer diagnosis were associated with increased mortality, particularly in patients with absent or low TIL.
Alexander FC Hulsbergen, Yu Tung Lo, Ilia Awakimjan, Vasileios K Kavouridis, John G Phillips, Timothy R Smith, Joost JC Verhoeff, Kun-Hsing Yu, Marike LD Broekman, and Omar Arnaout. 2022. “Survival Prediction After Neurosurgical Resection of Brain Metastases: A Machine Learning Approach.” Neurosurgery, 91, 3, Pp. 381-388.Abstract
BACKGROUND: Current prognostic models for brain metastases (BMs) have been constructed and validated almost entirely with data from patients receiving up-front radiotherapy, leaving uncertainty about surgical patients. OBJECTIVE: To build and validate a model predicting 6-month survival after BM resection using different machine learning algorithms. METHODS: An institutional database of 1062 patients who underwent resection for BM was split into an 80:20 training and testing set. Seven different machine learning algorithms were trained and assessed for performance; an established prognostic model for patients with BM undergoing radiotherapy, the diagnosis-specific graded prognostic assessment, was also evaluated. Model performance was assessed using area under the curve (AUC) and calibration. RESULTS: The logistic regression showed the best performance with an AUC of 0.71 in the hold-out test set, a calibration slope of 0.76, and a calibration intercept of 0.03. The diagnosis-specific graded prognostic assessment had an AUC of 0.66. Patients were stratified into regular-risk, high-risk and very high-risk groups for death at 6 months; these strata strongly predicted both 6-month and longitudinal overall survival ( P < .0005). The model was implemented into a web application that can be accessed through http://brainmets.morethanml.com . CONCLUSION: We developed and internally validated a prediction model that accurately predicts 6-month survival after neurosurgical resection for BM and allows for meaningful risk stratification. Future efforts should focus on external validation of our model.
Benjamin D Lee, Anthony Gitter, Casey S Greene, Sebastian Raschka, Finlay Maguire, Alexander J Titus, Michael D Kessler, Alexandra J Lee, Marc G Chevrette, Paul Allen Stewart, Thiago Britto-Borges, Evan M Cofer, Kun-Hsing Yu, Juan Jose Carmona, Elana J Fertig, Alexandr A Kalinin, Brandon Signal, Benjamin J Lengerich, Timothy J Triche, and Simina M Boca. 2022. “Ten quick tips for deep learning in biology.” PLoS Comput Biol, 18, 3, Pp. e1009803.
2021
Eliana Marostica, Rebecca Barber, Thomas Denize, Isaac S Kohane, Sabina Signoretti, Jeffrey A Golden, and Kun-Hsing Yu. 2021. “Development of a Histopathology Informatics Pipeline for Classification and Prediction of Clinical Outcomes in Subtypes of Renal Cell Carcinoma.” Clin Cancer Res, 27, 10, Pp. 2868-2878.Abstract
PURPOSE: Histopathology evaluation is the gold standard for diagnosing clear cell (ccRCC), papillary, and chromophobe renal cell carcinoma (RCC). However, interrater variability has been reported, and the whole-slide histopathology images likely contain underutilized biological signals predictive of genomic profiles. EXPERIMENTAL DESIGN: To address this knowledge gap, we obtained whole-slide histopathology images and demographic, genomic, and clinical data from The Cancer Genome Atlas, the Clinical Proteomic Tumor Analysis Consortium, and Brigham and Women's Hospital (Boston, MA) to develop computational methods for integrating data analyses. Leveraging these large and diverse datasets, we developed fully automated convolutional neural networks to diagnose renal cancers and connect quantitative pathology patterns with patients' genomic profiles and prognoses. RESULTS: Our deep convolutional neural networks successfully detected malignancy (AUC in the independent validation cohort: 0.964-0.985), diagnosed RCC histologic subtypes (independent validation AUCs of the best models: 0.953-0.993), and predicted stage I ccRCC patients' survival outcomes (log-rank test P = 0.02). Our machine learning approaches further identified histopathology image features indicative of copy-number alterations (AUC > 0.7 in multiple genes in patients with ccRCC) and tumor mutation burden. CONCLUSIONS: Our results suggest that convolutional neural networks can extract histologic signals predictive of patients' diagnoses, prognoses, and genomic variations of clinical importance. Our approaches can systematically identify previously unknown relations among diverse data modalities.
Shannon Wongvibulsin, Vartan Pahalyants, Mark Kalinich, William Murphy, Kun-Hsing Yu, Feicheng Wang, Steven T Chen, Kerry Reynolds, Shawn G Kwatra, and Yevgeniy R Semenov. 2021. “Epidemiology and risk factors for the development of cutaneous toxicities in patients treated with immune-checkpoint inhibitors: A United States population-level analysis.” J Am Acad Dermatol.Abstract
BACKGROUND: A variety of dermatoses have been reported in the growing number of patients treated with immune-checkpoint inhibitors (ICIs), but the current understanding of cutaneous immune-related adverse events (irAEs) is limited. OBJECTIVE: To determine the cumulative incidence, distribution, and risk factors of cutaneous irAEs after ICI initiation. METHODS: This was a retrospective cohort study of patients in a national insurance claims database including cancer patients treated with ICIs and matched controls. RESULTS: The study included 8637 ICI patients and 8637 matched controls. The overall incidence of cutaneous irAEs was 25.1%, with a median onset time of 113 days. The ICI group had a significantly higher incidence of pruritus, mucositis, erythroderma, maculopapular eruption, vitiligo, lichen planus, bullous pemphigoid, Grover disease, rash, other nonspecific eruptions, and drug eruption or other nonspecific drug reaction. Patients with melanoma and renal cell carcinoma and those receiving combination therapy were at a higher risk of cutaneous irAEs. LIMITATIONS: Retrospective design without access to patient chart data. CONCLUSIONS: This study identifies cutaneous irAEs in a real-world clinical setting and highlights patient groups that are particularly at risk. The results can aid dermatologists at the bedside in the diagnosis of cutaneous irAEs and in formulating management recommendations to referring oncologists regarding the continuation of ICI therapy.
Yasha Ektefaie, William Yuan, Deborah A Dillon, Nancy U Lin, Jeffrey A Golden, Isaac S Kohane, and Kun-Hsing Yu. 2021. “Integrative multiomics-histopathology analysis for breast cancer classification.” NPJ Breast Cancer, 7, 1, Pp. 147.Abstract
Histopathologic evaluation of biopsy slides is a critical step in diagnosing and subtyping breast cancers. However, the connections between histology and multi-omics status have never been systematically explored or interpreted. We developed weakly supervised deep learning models over hematoxylin-and-eosin-stained slides to examine the relations between visual morphological signal, clinical subtyping, gene expression, and mutation status in breast cancer. We first designed fully automated models for tumor detection and pathology subtype classification, with the results validated in independent cohorts (area under the receiver operating characteristic curve ≥ 0.950). Using only visual information, our models achieved strong predictive performance in estrogen/progesterone/HER2 receptor status, PAM50 status, and TP53 mutation status. We demonstrated that these models learned lymphocyte-specific morphological signals to identify estrogen receptor status. Examination of the PAM50 cohort revealed a subset of PAM50 genes whose expression reflects cancer morphology. This work demonstrates the utility of deep learning-based image models in both clinical and research regimes, through its ability to uncover connections between visual morphology and genetic statuses.
Mark Kalinich, William Murphy, Shannon Wongvibulsin, Vartan Pahalyants, Kun-Hsing Yu, Chenyue Lu, Feicheng Wang, Leyre Zubiri, Vivek Naranbhai, Alexander Gusev, Shawn G Kwatra, Kerry L Reynolds, and Yevgeniy R Semenov. 2021. “Prediction of severe immune-related adverse events requiring hospital admission in patients on immune checkpoint inhibitors: study of a population level insurance claims database from the USA.” J Immunother Cancer, 9, 3.Abstract
BACKGROUND: Immune-related adverse events (irAEs) are a serious side effect of immune checkpoint inhibitor (ICI) therapy for patients with advanced cancer. Currently, predisposing risk factors are undefined but understanding which patients are at increased risk for irAEs severe enough to require hospitalization would be beneficial to tailor treatment selection and monitoring. METHODS: We performed a retrospective review of patients with cancer treated with ICIs using unidentifiable claims data from an Aetna nationwide US health insurance database from January 3, 2011 to December 31, 2019, including patients with an identified primary cancer and at least one administration of an ICI. Regression analyses were performed. Main outcomes were incidence of and factors associated with irAE requiring hospitalization in ICI therapy. RESULTS: There were 68.8 million patients identified in the national database, and 14 378 patients with cancer identified with at least 1 administration of ICI in the study period. Patients were followed over 19 117 patient years and 504 (3.5%) developed an irAE requiring hospitalization. The incidence of irAEs requiring hospitalization per patient ICI treatment year was 2.6%, rising from 0% (0/71) in 2011 to 3.7% (93/2486) in 2016. Combination immunotherapy (OR: 2.44, p<0.001) was associated with increased odds of developing irAEs requiring hospitalization, whereas older patients (OR 0.98 per additional year, p<0.001) and those with non-lung cancer were associated with decreased odds of irAEs requiring hospitalization (melanoma OR: 0.70, p=0.01, renal cell carcinoma OR: 0.71, p=0.03, other cancers OR: 0.50, p<0.001). Sex, region, zip-code-imputed income, and zip-code unemployment were not associated with incidence of irAE requiring hospitalization. Prednisone (72%) and methylprednisolone (25%) were the most common immunosuppressive treatments identified in irAE hospitalizations. CONCLUSIONS: We found that 3.5% of patients initiating ICI therapy experienced irAEs requiring hospitalization and immunosuppression. The odds of irAEs requiring hospitalization were higher with younger age, treatment with combination ICI therapy (cytotoxic T lymphocyte-associated 4 and programmed cell death protein 1 (PD-1) or programmed death-ligand 1 (PD-L1)), and lower for other cancers compared with patients on PD-1 or PD-L1 inhibitors with lung cancer. This evidence from the first nationwide study of irAEs requiring hospitalization in the USA identified the real-world epidemiology, risk factors, and treatment patterns of these irAEs which may guide treatment and management decisions.
Oren Miron, Rafael E Delgado, Christine F Delgado, Elizabeth A Simpson, Kun-Hsing Yu, Anibal Gutierrez, Guangyu Zeng, Jillian N Gerstenberger, and Isaac S Kohane. 2021. “Prolonged Auditory Brainstem Response in Universal Hearing Screening of Newborns with Autism Spectrum Disorder.” Autism Res, 14, 1, Pp. 46-52.Abstract
Previous studies report prolonged auditory brainstem response (ABR) in children and adults with autism spectrum disorder (ASD). Despite its promise as a biomarker, it is unclear whether healthy newborns who later develop ASD also show ABR abnormalities. In the current study, we extracted ABR data on 139,154 newborns from their Universal Newborn Hearing Screening, including 321 newborns who were later diagnosed with ASD. We found that the ASD newborns had significant prolongations of their ABR phase and V-negative latency compared with the non-ASD newborns. Newborns in the ASD group also exhibited greater variance in their latencies compared to previous studies in older ASD samples, likely due in part to the low intensity of the ABR stimulus. These findings suggest that newborns display neurophysiological variation associated with ASD at birth. Future studies with higher-intensity stimulus ABRs may allow more accurate predictions of ASD risk, which could augment the universal ABR test that currently screens millions of newborns worldwide. LAY SUMMARY: Children with autism spectrum disorder (ASD) have slow brain responses to sounds. We examined these brain responses from newborns' hearing tests and found that newborns who were later diagnosed with autism also had slower brain responses to sounds. Future studies might use these findings to better predict autism risk, with a hearing test that is already used on millions of newborns worldwide.
Feicheng Wang, Shihao Yang, Nathan Palmer, Kathe Fox, Isaac S Kohane, Katherine P Liao, Kun-Hsing Yu, and SC Kou. 2021. “Real-world data analyses unveiled the immune-related adverse effects of immune checkpoint inhibitors across cancer types.” NPJ Precis Oncol, 5, 1, Pp. 82.Abstract
Immune checkpoint inhibitors have demonstrated significant survival benefits in treating many types of cancers. However, their immune-related adverse events (irAEs) have not been systematically evaluated across cancer types in large-scale real-world populations. To address this gap, we conducted real-world data analyses using nationwide insurance claims data with 85.97 million enrollees across 8 years. We identified a significantly increased risk of developing irAEs among patients receiving immunotherapy agents in all seven cancer types commonly treated with immune checkpoint inhibitors. By six months after treatment initialization, those receiving immunotherapy were 1.50-4.00 times (95% CI, lower bound from 1.15 to 2.16, upper bound from 1.69 to 20.36) more likely to develop irAEs in the first 6 months of treatment, compared to matched chemotherapy or targeted therapy groups, with a total of 92,858 patients. The risk of developing irAEs among patients using nivolumab is higher compared to those using pembrolizumab. These results confirmed the need for clinicians to assess irAEs among cancer patients undergoing immunotherapy as part of management. Our methods are extensible to characterizing the effectiveness and adverse effects of novel treatments in large populations in an efficient and economical fashion.
William Yuan, Brett K Beaulieu-Jones, Kun-Hsing Yu, Scott L Lipnick, Nathan Palmer, Joseph Loscalzo, Tianxi Cai, and Isaac S Kohane. 2021. “Temporal bias in case-control design: preventing reliable predictions of the future.” Nat Commun, 12, 1, Pp. 1107.Abstract
One of the primary tools that researchers use to predict risk is the case-control study. We identify a flaw, temporal bias, that is specific to and uniquely associated with these studies that occurs when the study period is not representative of the data that clinicians have during the diagnostic process. Temporal bias acts to undermine the validity of predictions by over-emphasizing features close to the outcome of interest. We examine the impact of temporal bias across the medical literature, and highlight examples of exaggerated effect sizes, false-negative predictions, and replication failure. Given the ubiquity and practical advantages of case-control studies, we discuss strategies for estimating the influence of and preventing temporal bias where it exists.
2020
Kun-Hsing Yu, Feiran Wang, Gerald J Berry, Christopher Ré, Russ B Altman, Michael Snyder, and Isaac S Kohane. 2020. “Classifying non-small cell lung cancer types and transcriptomic subtypes using convolutional neural networks.” J Am Med Inform Assoc, 27, 5, Pp. 757-769.Abstract
OBJECTIVE: Non-small cell lung cancer is a leading cause of cancer death worldwide, and histopathological evaluation plays the primary role in its diagnosis. However, the morphological patterns associated with the molecular subtypes have not been systematically studied. To bridge this gap, we developed a quantitative histopathology analytic framework to identify the types and gene expression subtypes of non-small cell lung cancer objectively. MATERIALS AND METHODS: We processed whole-slide histopathology images of lung adenocarcinoma (n = 427) and lung squamous cell carcinoma patients (n = 457) in the Cancer Genome Atlas. We built convolutional neural networks to classify histopathology images, evaluated their performance by the areas under the receiver-operating characteristic curves (AUCs), and validated the results in an independent cohort (n = 125). RESULTS: To establish neural networks for quantitative image analyses, we first built convolutional neural network models to identify tumor regions from adjacent dense benign tissues (AUCs > 0.935) and recapitulated expert pathologists' diagnosis (AUCs > 0.877), with the results validated in an independent cohort (AUCs = 0.726-0.864). We further demonstrated that quantitative histopathology morphology features identified the major transcriptomic subtypes of both adenocarcinoma and squamous cell carcinoma (P < .01). DISCUSSION: Our study is the first to classify the transcriptomic subtypes of non-small cell lung cancer using fully automated machine learning methods. Our approach does not rely on prior pathology knowledge and can discover novel clinically relevant histopathology patterns objectively. The developed procedure is generalizable to other tumor types or diseases.
Oren Miron, Kun-Hsing Yu, Rachel Wilf-Miron, Isaac Kohane, and Nadav Davidovitch. 2020. “COVID-19 infections following physical school reopening.” Arch Dis Child.
Kun-Hsing Yu, Vincent Hu, Feiran Wang, Ursula A Matulonis, George L Mutter, Jeffrey A Golden, and Isaac S Kohane. 2020. “Deciphering serous ovarian carcinoma histopathology and platinum response by convolutional neural networks.” BMC Med, 18, 1, Pp. 236.Abstract
BACKGROUND: Ovarian cancer causes 151,900 deaths per year worldwide. Treatment and prognosis are primarily determined by the histopathologic interpretation in combination with molecular diagnosis. However, the relationship between histopathology patterns and molecular alterations is not fully understood, and it is difficult to predict patients' chemotherapy response using the known clinical and histological variables. METHODS: We analyzed the whole-slide histopathology images, RNA-Seq, and proteomics data from 587 primary serous ovarian adenocarcinoma patients and developed a systematic algorithm to integrate histopathology and functional omics findings and to predict patients' response to platinum-based chemotherapy. RESULTS: Our convolutional neural networks identified the cancerous regions with areas under the receiver operating characteristic curve (AUCs) > 0.95 and classified tumor grade with AUCs > 0.80. Functional omics analysis revealed that expression levels of proteins participated in innate immune responses and catabolic pathways are associated with tumor grade. Quantitative histopathology analysis successfully stratified patients with different response to platinum-based chemotherapy (P = 0.003). CONCLUSIONS: These results indicated the potential clinical utility of quantitative histopathology evaluation in tumor cell detection and chemotherapy response prediction. The developed algorithm is easily extensible to other tumor types and treatment modalities.
Eliana Marostica and Kun-Hsing Yu. 2020. “Deep decision support for lymph node metastatic risk evaluation.” EBioMedicine, 62, Pp. 103105.
Iván Sánchez Fernández, Edward Yang, Paola Calvachi, Marta Amengual-Gual, Joyce Y Wu, Darcy Krueger, Hope Northrup, Martina E Bebin, Mustafa Sahin, Kun-Hsing Yu, and Jurriaan M Peters. 2020. “Deep learning in rare disease. Detection of tubers in tuberous sclerosis complex.” PLoS One, 15, 4, Pp. e0232376.Abstract
OBJECTIVE: To develop and test a deep learning algorithm to automatically detect cortical tubers in magnetic resonance imaging (MRI), to explore the utility of deep learning in rare disorders with limited data, and to generate an open-access deep learning standalone application. METHODS: T2 and FLAIR axial images with and without tubers were extracted from MRIs of patients with tuberous sclerosis complex (TSC) and controls, respectively. We trained three different convolutional neural network (CNN) architectures on a training dataset and selected the one with the lowest binary cross-entropy loss in the validation dataset, which was evaluated on the testing dataset. We visualized image regions most relevant for classification with gradient-weighted class activation maps (Grad-CAM) and saliency maps. RESULTS: 114 patients with TSC and 114 controls were divided into a training set, a validation set, and a testing set. The InceptionV3 CNN architecture performed best in the validation set and was evaluated in the testing set with the following results: sensitivity: 0.95, specificity: 0.95, positive predictive value: 0.94, negative predictive value: 0.95, F1-score: 0.95, accuracy: 0.95, and area under the curve: 0.99. Grad-CAM and saliency maps showed that tubers resided in regions most relevant for image classification within each image. A stand-alone trained deep learning App was able to classify images using local computers with various operating systems. CONCLUSION: This study shows that deep learning algorithms are able to detect tubers in selected MRI images, and deep learning can be prudently applied clinically to manually selected data in a rare neurological disorder.

Pages