Automated curation of large-scale cancer histopathology image datasets using deep learning.

BACKGROUND: Artificial intelligence (AI) has numerous applications in pathology, supporting diagnosis and prognostication in cancer. However, most AI models are trained on highly selected data, typically one tissue slide per patient. In reality, especially for large surgical resection specimens, dozens of slides can be available for each patient. Manually sorting and labelling whole-slide images (WSIs) is a very time-consuming process, hindering the direct application of AI on the collected tissue samples from large cohorts. In this study we addressed this issue by developing a deep-learning (DL)-based method for automatic curation of large pathology datasets with several slides per patient. METHODS: We collected multiple large multicentric datasets of colorectal cancer histopathological slides from the United Kingdom (FOXTROT, N = 21,384 slides; CR07, N = 7985 slides) and Germany (DACHS, N = 3606 slides). These datasets contained multiple types of tissue slides, including bowel resection specimens, endoscopic biopsies, lymph node resections, immunohistochemistry-stained slides, and tissue microarrays. We developed, trained, and tested a deep convolutional neural network model to predict the type of slide from the slide overview (thumbnail) image. The primary statistical endpoint was the macro-averaged area under the receiver operating curve (AUROCs) for detection of the type of slide. RESULTS: In the primary dataset (FOXTROT), with an AUROC of 0.995 [95% confidence interval [CI]: 0.994-0.996] the algorithm achieved a high classification performance and was able to accurately predict the type of slide from the thumbnail image alone. In the two external test cohorts (CR07, DACHS) AUROCs of 0.982 [95% CI: 0.979-0.985] and 0.875 [95% CI: 0.864-0.887] were observed, which indicates the generalizability of the trained model on unseen datasets. With a confidence threshold of 0.95, the model reached an accuracy of 94.6% (7331 classified cases) in CR07 and 85.1% (2752 classified cases) for the DACHS cohort. CONCLUSION: Our findings show that using the low-resolution thumbnail image is sufficient to accurately classify the type of slide in digital pathology. This can support researchers to make the vast resource of existing pathology archives accessible to modern AI models with only minimal manual annotations.

Factors associated with immunological non-response after ART initiation: a retrospective observational cohort study.

BACKGROUND: Among people living with HIV (PLHIV) on antiretroviral therapy (ART), the mortality of immunological non-responders (INRs) is higher than that of immunological responders (IRs). However, factors associated with immunological non-response following ART are not well documented. METHODS: We obtained data for HIV patients from the National Free Antiretroviral Treatment Program database in China. Patients were grouped into IRs (CD4 cell count ≥ 350 cells/µl after 24 months' treatment), immunological incomplete responders (ICRs) (200-350 cells/µl) and INRs (< 200 cells/µl). Multivariable logistic regression was used to assess factors associated with immunological non-response. RESULTS: A total of 3900 PLHIV were included, among whom 2309 (59.2%) were IRs, 1206 (30.9%) ICRs and 385 (9.9%) INRs. In multivariable analysis, immunological non-response was associated with being male (2.07, 1.39-3.09), older age [40-49 years (vs. 18-29 years): 2.05, 1.29-3.25; 50-59 years: 4.04, 2.33-7.00; ≥ 60 years: 5.51, 2.84-10.67], HBV co-infection (1.63, 1.14-2.34), HCV co-infection (2.01, 1.01-4.02), lower CD4 + T cell count [50-200 cells/µl (vs. 200-350 cells/µl): 40.20, 16.83-96.01; < 50 cells/µl: 215.67, 85.62-543.26] and lower CD4/CD8 ratio (2.93, 1.98-4.34) at baseline. Compared with patients treated with non-nucleoside reverse transcriptase inhibitors (NNRTIs) based regimens, those receiving protease inhibitors (PIs) based regimens were less likely to be INRs (0.47, 0.26-0.82). CONCLUSIONS: We found a sizable immunological non-response rate among HIV-infected patients. Being male, older age, coinfection with HBV and HCV, lower CD4 + T cell count and lower CD4/CD8 ratio are risk factors of immunological non-response, whereas PIs-based regimens is a protective factor.

Acceptability of male circumcision for HIV prevention among men who have sex with men in China: a short report.

Recent evidence shows that circumcision is associated with lower HIV prevalence among MSM. We assessed the acceptability of circumcision for preventing HIV and that of Shang Ring circumcision (SRC) among men who have sex with men (MSM) in China. 538 adult MSM were recruited from six cities in China between January and March 2019. Participants were surveyed by an online, self-administered questionnaire. The acceptability of circumcision was assessed before and after the potential protective effect of circumcision against HIV was informed, and subsequently men's willingness to undergo SRC was assessed. The level of circumcision was 16.4%. Of 450 uncircumcised MSM, their willingness to be circumcised in the following six months increased significantly from 32.2% to 55.6% after the information session. Three quarters of men who were willing to undergo circumcision accepted SRC. MSM who perceived that circumcision could maintain genital hygiene were more likely to accept circumcision after the information session, whereas those who regarded circumcision as an embarrassing surgery were disinclined to be circumcised. The low circumcision rate, along with its high acceptability in Chinese MSM, suggests a great potential benefit of circumcision intervention if proved effective. SRC might be a popular circumcision procedure in this population.

Outness and its correlates among gay and bisexual men attending university in China: a cross-sectional study.

BACKGROUND: Disclosure of sexual orientation to others (outness) might be associated with sexual and mental health among gay and bisexual men (GBM) attending university. We aimed to characterise outness and investigate factors correlated with outness among GBM attending university in China. METHODS: Between September 2018 and March 2019, GBM attending university were recruited in six cities in China. Information on sociodemographic characteristics, outness and sexual behaviours were collected using a self-administered questionnaire. Each participant was tested for HIV/STIs. Correlates of outness were assessed using multivariable logistic regression. RESULTS: A total of 400 GBM attending university were recruited, of whom 251 (62.8%) had disclosed their sexual orientation. Men who served as student leaders (adjusted odds ratio [AOR]=2.28, 95% CI: 1.46-3.54) and donated blood (AOR 1.85, 95% CI: 1.05-3.24) were more likely to disclose their sexual orientation, whereas men who had sex with a female (AOR 0.19, 95% CI: 0.05-0.74) and had group sex (AOR 0.52, 95% CI: 0.30-0.89) were less likely to disclose their sexual orientation. Mental health status, HIV/STI infections were not associated with outness. CONCLUSIONS: GBM attending university who disclosed their sexual orientation were more likely to be involved with student work and less likely to engage in high-risk sexual behaviours. More attention and education could focus on non-disclosing GBM men attending university through peer education or other ways.

Awareness and uptake of non-occupational post-exposure prophylaxis (nPEP) among gay, bisexual and other men who have sex with men in China: a cross-sectional study.

Non-occupational post-exposure prophylaxis (nPEP) is recommended to reduce the likelihood of HIV infection after potential exposure. However, little is known about this practice among Chinese gay, bisexual and other men who have sex with men (GBM). GBM were enrolled from both centers for disease prevention and control (CDCs) and community health centers in six cities in China. Multivariable logistics regression was used to assess factors associated with awareness of and willingness to use nPEP. A total of 516 eligible GBM were included, 67.2% of whom were aware of nPEP, 76.0% were willing to use nPEP, and 2.3% had ever used nPEP. GBM who had college or higher education, had disclosed sexual orientation, had increased number of male sex partners in the last 6 months, and had ever tested for HIV were more likely to be aware of nPEP. Willingness to use nPEP was significantly associated with college or higher education, STI history, gay app use, and awareness of nPEP prior to study. Uptake of nPEP is still low among MSM in China. Efforts are needed to improve awareness of and access to nPEP among GBM, especially those at higher risk of HIV infection.

Rectal Douching and Anal Human Papillomavirus Among Young Men Who Have Sex With Men in China.

BACKGROUND: Young men who have sex with men (YMSM) are particularly at increased risk for human papillomavirus (HPV) infection and related malignancies. Rectal douching, commonly practiced among MSM, has been associated with HIV, chlamydia, gonorrhea, and viral hepatitis in cross-sectional studies. We conducted this study to understand the association between self-reported rectal douching and anal HPV among YMSM in China. METHODS: Between September 2018 and March 2019, MSM aged 15 to 24 years who ever engaged in receptive anal intercourse over the last 3 months were recruited via community-based organizations and centers for disease control located in 4 cities in China. Participants were asked to complete an online survey. We performed multivariate logistic regression adjusted for potential confounders to examine the association between self-reported rectal douching and anal HPV. RESULTS: Among 273 MSM with a median age of 20 years (interquartile range, 19-21 years) included in this study, 130 (47.6%) practiced rectal douching during the preceding 3 months and 96 (36.2%) were infected with anal HPV. Self-reported rectal douching was found to be associated with increased odds of anal HPV infection (odds ratio, 2.16; 95% confidence interval, 1.22-3.82) among YMSM, after adjusting for age, sexual debut, sex with alcohol or drugs, and HIV testing history. CONCLUSIONS: Self-reported rectal douching is associated with higher odds of anal HPV infection independent of sexual behaviors among YMSM. More prospective studies to ascertain this association are needed. Health education materials should inform men of the potential risk of rectal douching. Research on safer rectal douching procedures is warranted.

Human papillomavirus prevalence among men who have sex with men in China: a systematic review and meta-analysis.

Human papillomavirus (HPV) infection among men who have sex with men (MSM) in China is underreported. We performed a systematic review and meta-analysis to clarify site-specific HPV prevalence among MSM in China. We searched both English and Chinese databases for all studies published before April 1, 2020, that reported HPV prevalence among MSM in China. Random-effects meta-analysis was used to calculate summary estimates. Thirty-four articles were eligible, where 32, 5, and 2 articles reported HPV prevalence at the anus, penis, and oral cavity, respectively. The estimated prevalence of anal HPV among MSM in China was 85.1% (HIV-positive), 53.6% (HIV-negative), and 59.2% (unknown HIV status), with HPV genotypes being predominated by HPV 6, 11, 16, 18, 52, and 58. Any HPV and high-risk (HR) HPV was more common in northern China, while low-risk HPV was more common in southern China. HPV prevalence increased with age among HIV-negative MSM, from 40.5% (aged < 20 years) to 57.2% (aged ≥ 40 years). High prevalence of any HPV (HIV+: 95.1%; HIV-: 97.7%) and multiple infections (HIV+: 75.9%; HIV-: 41.7%) was found in anogenital warts among MSM. HPV is common among MSM in China. MSM living with HIV and/or anogenital warts were at disproportionate risk for HR HPV. Younger MSM were found to have a lower HPV prevalence. HPV vaccines would have prevented the majority of infections if given before sex debut. HPV at anatomical sites other than the anus, incident HPV infection, and the cost-effectiveness of HPV vaccination in this population are worth further investigation.

Effect of availability of HIV self-testing on HIV testing frequency among men who have sex with men attending university in China (UniTest): protocol of a stepped-wedge randomized controlled trial.

BACKGROUND: HIV testing plays a central role in the combat against HIV. We aimed to determine if the availability of HIV self-testing (HIVST) would increase the frequency of testing among men who have sex with men (MSM) attending university in China. METHODS: A stepped wedge randomized controlled trial will be conducted in 4 provinces in China: Chongqing, Guangdong, Shandong, and Tianjin. Eligibility assessment will include (1) male, aged 16 years or older, (2) university student (technical diploma and undergraduate students), (3) MSM (sexual behaviors including mutual masturbation, oral sex, and anal sex), (4) HIV negative, and (5) willing to provide informed consent. Participants will be randomly allocated to HIV self-testing intervention with free HIVST kits in every 30 days according to the intervention waiting lists with a computer-generated randomized sequence. All participants will complete a self-administrated online questionnaire onsite at baseline and 12-month follow-up and complete an online questionnaire at 4- and 8-month. The primary outcome is the effect of HIVST on HIV testing frequency. Secondary outcomes include the change in sexual behaviors and HIV incidence. DISCUSSION: No previous study had measured the effect of social media based HIVST intervention on the change in HIV testing behaviors, sexual behaviors and incident HIV infection among MSM attending university in China. Findings from this study will provide evidence for further interventional practice promotions and prevention strategies scale-up, including HIV testing, pre-exposure prophylaxis (PrEP) or post-exposure prophylaxis (PEP), and sexual partner serosorting. TRIAL REGISTRATION: Chinese Clinical Trial Registry: ChiCTR1900020645. Registered 11 January 2019.

Efficacy of HIV interventions among factory workers in low- and middle-income countries: a systematic review.

BACKGROUND: Factory workers in low- and middle-income countries (LMICs) are vulnerable to HIV transmission. Interventions are needed to prevent HIV in this population. We systematically reviewed published literature on the efficacy of various HIV interventions in reducing stigma, risk behaviors and HIV transmission among factory workers. METHODS: A systematic review was performed using predefined inclusion and exclusion criteria. Four databases (PubMed, PsycINFO, Scopus and EMBASE) were searched for relevant publications between January 1, 1990 and December 31, 2018. Two independent reviewers assessed the methodological quality of studies. RESULTS: Thirteen articles were included, with 2 randomized controlled trials and 11 cohort studies. Five interventions and their combinations were summarized. Educational intervention increased condom use and reduced the use of recreational drugs and alcohol before sex. Community intervention that proactively provide HIV counselling and testing (HCT) services could increase the detection rate of HIV and other sexually transmitted diseases (STDs). Lottery intervention increased HCT uptake and decreased HIV public stigma. Education combined with community intervention reduced the proportion of workers with casual sex and enhanced HIV knowledge. Peer education combined with community intervention increased the proportion of workers who were willing to take their partners to HCT. Policy intervention combined with peer education enhanced HIV knowledge, perceived condom accessibility and condom use with regular partners. CONCLUSIONS: Various interventions improved HIV knowledge, decreased HIV stigma and reduced HIV-related risk behaviors among factory workers in LMICs. The combination of multiple interventions tended to achieve better efficacy than a single intervention. Persistent combination interventions are essential to address HIV in this population.

Rectal douching among men who have sex with men in six cities in China: a cross-sectional study.

Background Rectal douching, which is a common practice among men who have sex men (MSM) as a preparation for anal sex, may potentially increase the risk of HIV infection. However, little is known about the practice of rectal douching among Chinese MSM. This study investigated the characteristics of rectal douching and factors associated with rectal douching. METHODS: Between January and March 2019, MSM aged ≥18 years in six cities in China were recruited to the study. Participants were asked to complete a self-administered online questionnaire. Chi-squared tests and multivariable logistic regression were used to examine factors associated with rectal douching by calculating odds ratios (ORs) and 95% confidence intervals (CIs). RESULTS: Of 485 MSM in this study, 278 (57.3%) practiced rectal douching in the preceding 6 months. The most common reasons for douching before and after anal sex were hygiene/cleanliness (78.6% and 66.1% respectively) and HIV/STI prevention (8.5% and 17.4% respectively). The most used enemas and equipment before and after anal sex were tap water (73.5% and 70.2% respectively) and showerhead hoses (76.1% and 75.2% respectively). Men who engaged in both insertive and receptive anal sex (OR 8.84; 95% CI 4.52-17.30), exclusively receptive anal sex (OR 6.56; 95% CI 3.55-12.13), condom-less anal intercourse (OR 1.64; 95% CI 1.03-2.63) and anal sex after alcohol use (OR 1.54; 95% CI 1.00-2.35) were more likely to perform douching. CONCLUSION: Rectal douching is common in MSM, some of whom may not realise the potential risks associated with douching. The co-occurrence of douching and other high-risk behaviours would make MSM more vulnerable to HIV. Health education aimed at informing MSM of the potential risks and addressing the high prevalence of douching are urgently needed.

Use of gay app and the associated HIV/syphilis risk among non-commercial men who have sex with men in Shenzhen, China: a serial cross-sectional study.

OBJECTIVES: Geosocial networking application specific to men who have sex with men (MSM) (gay app) has revolutionised the social networking of MSM globally, much concern was raised over its linkage to HIV/syphilis risk. This study sought to examine the association between use of gay app and sexual behaviours and HIV/syphilis risk among Chinese MSM. METHODS: Eligible MSM were recruited through combined offline methods from 2015 to 2017 in Shenzhen, China, with data collected including demographics, sexual behaviours, app use, recreational drug use and HIV testing. All participants are required to sign a written informed consent and take a confidential HIV and syphilis testing. RESULTS: The prevalence of app use among non-commercial MSM (NcMSM) has rapidly increased from 12.5% in 2015 to 52.6% in 2017. The primary four apps used were Blued (97.2%), Aloha (18.4%), Jack'd (14.1) and Zank (14.1%). After controlling for confounders, HIV prevalence was still significantly lower among app users than non-app users (adjusted odds ratios [AOR]: 0.77, 95% CI 0.61 to 0.97), yet the lower prevalence of syphilis was not significant (AOR: 0.97, 95% CI 0.76 to 1.23). App-using NcMSM were more likely to be younger, unmarried, self-identified as homosexuality and having higher education level and income than non-app-using NcMSM. App-using NcMSM had higher rate of consistent condom use and HIV testing, higher level of knowledge on HIV/AIDS prevention and condom use; however, they were more likely to have multiple sexual partners, practice receptive role in anal sex and use recreational drug. CONCLUSIONS: App-using NcMSM are more likely to have sexual risk behaviours as well as risk-reduction behaviours such as consistent condom use and HIV testing. Scaled-up and innovative venue-based HIV interventions are warranted for these high-risk MSM frequent social venues with less condom use and fewer HIV tests. Meanwhile, gay app should alternatively serve as an intervention and education platform for the MSM hard-to-reach via venue-based approaches.

Association between rectal douching and HIV and other sexually transmitted infections among men who have sex with men: a systematic review and meta-analysis.

BACKGROUND: Men who have sex with men (MSM) are disproportionately affected by HIV and other STIs worldwide. Rectal douching, which is commonly used by MSM in preparation for anal sex, may increase the risk of HIV and other STIs by injuring the rectal mucosa. Results from individual studies reporting associations between rectal douching and HIV and other STIs among MSM are inconsistent. We performed a systematic review and meta-analysis to estimate the association between rectal douching and HIV and other STIs among MSM. METHODS: We searched PubMed, Embase, Scopus and Web of Science for studies published from January 1970 to November 2018. Studies that reported ORs and 95% CIs of associations between rectal douching and infection with HIV/STIs, or reported enough data to calculate these estimates, were included. We assessed risk of bias using the Newcastle-Ottawa Scale. ORs were pooled using a random effects model. RESULTS: Twenty-eight eligible studies were identified in our review, of which 24 (20 398 participants) were included in the meta-analysis. Rectal douching was associated with increased odds of infection with HIV (OR 2.80, 95% CI 2.32 to 3.39), and any STI other than HIV (including hepatitis B virus (HBV), hepatitis C virus (HCV), chlamydia, gonorrhoea, syphilis and human papillomavirus) (OR 2.46, 95% CI 1.95 to 3.11) among MSM. For specific STIs, douching was associated with increased odds of viral hepatitis (HBV, HCV) (OR 3.29, 95% CI 2.79 to 3.87), and chlamydia or gonorrhoea (OR 3.25, 95% CI 2.02 to 5.23). These associations remained significant in studies that adjusted for potential confounders. CONCLUSION: Rectal douching may put MSM at increased risk for infection with HIV and other STIs. Longitudinal studies are needed to clarify this association, and health education materials should inform men of the potential for increased risk of infection with rectal douching.

Clinical significance of combined tumour-infiltrating lymphocytes and microsatellite instability status in colorectal cancer: a systematic review and network meta-analysis.

BACKGROUND: Microsatellite instability (MSI) status and tumour-infiltrating lymphocytes (TIL) are established prognostic factors in colorectal cancer. Previous studies evaluating the combination of TIL and MSI status identified distinct colorectal cancer subtypes with unique prognostic associations. However, these studies were often limited by sample size, particularly for MSI-high (MSI-H) tumours, and there is no comprehensive summary of the available evidence. We aimed to review the literature to compare the survival outcomes associated with the subtypes derived from the integrated MSI-TIL classification in patients with colorectal cancer. METHODS: In this systematic review and network meta-analysis, we searched PubMed, Embase, Scopus, and the Cochrane Library without language restrictions, for articles published between Jan 1, 1990, and March 13, 2024. Patient cohorts comparing different combinations of TIL (high or low) and MSI status (MSI or microsatellite stable [MSS]) in patients with surgically resected colorectal cancer were included. Studies were excluded if they focused on neoadjuvant therapy or on other immune markers such as B cells or macrophages. Methodological quality assessment was done with the Newcastle-Ottawa scale; data appraisal and extraction was done independently by two reviewers. Summary estimates were extracted from published reports. The primary outcomes were overall survival, disease-free survival, and cancer-specific survival. A frequentist network meta-analysis was done to compare hazard ratios (HRs) and 95% CI for each outcome. The MSI-TIL subgroups were prognostically ranked based on P-score, bias, magnitude, and precision of associations with each outcome. The protocol is registered with PROSPERO (CRD42023461108). FINDINGS: Of 302 studies initially identified, 21 studies (comprising 14 028 patients) were included in the systematic review and 19 (13 029 patients) in the meta-analysis. Nine studies were identified with a low risk of bias and the remaining ten had a moderate risk of bias. The MSI-TIL-high (MSI-TIL-H) subtype exhibited longer overall survival (HR 0·45, 95% CI 0·34-0·61; I2=77·7%), disease-free survival (0·43, 0·32-0·58; I2=61·6%), and cancer-specific survival (0·53, 0·43-0·66; I2=0%), followed by the MSS-TIL-H subtype for overall survival (HR 0·53, 0·41-0·69; I2=77·7%), disease-free survival (0·52, 0·41-0·64; I2=61·6%), and cancer-specific survival (0·55, 0·47-0·64; I2=0%) than did patients with MSS-TIL-low tumours (MSS-TIL-L). Patients with the MSI-TIL-L subtype had similar overall survival (0·88, 0·66-1·18; I2=77·7%) and disease-free survival (0·93, 0·69-1·26; I2=61·6%), but a modestly longer cancer-specific survival (0·72, 0·57-0·90; I2=0%) than did the MSS-TIL-L subtype. Results from the direct and indirect evidence were strongly congruous. INTERPRETATION: The findings from this network meta-analysis suggest that better survival was only observed among patients with TIL-H colorectal cancer, regardless of MSI or MSS status. The integrated MSI-TIL classification should be further explored as a predictive tool for clinical decision-making in early-stage colorectal cancer. FUNDING: German Research Council (HO 5117/2-2).

CpG-biomarkers in tumor tissue and prediction models for the survival of colorectal cancer: A systematic review and external validation study.

The research aimed to identify previously published CpG-methylation-based prognostic biomarkers and prediction models for colorectal cancer (CRC) prognosis and validate them in a large external cohort. A systematic search was conducted, analyzing 298 unique CpGs and 12 CpG-based prognostic models from 28 studies. After adjustment for clinical variables, 48 CpGs and five prognostic models were confirmed to be associated with survival. However, the discrimination ability of the models was insufficient, with area under the receiver operating characteristic curves ranging from 0.53 to 0.62. Calibration accuracy was mostly poor, and no significant added prognostic value beyond traditional clinical variables was observed. All prognostic models were rated at high risk of bias. While a fraction of CpGs showed potential clinical utility and generalizability, the CpG-based prognostic models performed poorly and lacked clinical relevance.

Large-scale external validation and meta-analysis of gene methylation biomarkers in tumor tissue for colorectal cancer prognosis.

BACKGROUND: DNA methylation biomarkers in colorectal cancer (CRC) tissue hold potential as prognostic indicators. However, individual studies have yielded heterogeneous results, and external validation is largely absent. We conducted a comprehensive external validation and meta-analysis of previously suggested gene methylation biomarkers for CRC prognosis. METHODS: We performed a systematic search to identify relevant studies investigating gene methylation biomarkers for CRC prognosis until March 2024. Our external validation cohort with long-term follow-up included 2303 patients with CRC from 22 hospitals in southwest Germany. We used Cox regression analyses to assess associations between previously suggested gene methylation biomarkers and prognosis, adjusting for clinical variables. We calculated pooled hazard ratios (HRs) and their 95% confidence intervals (CIs) using random-effects models. FINDINGS: Of 151 single gene and 29 multiple gene methylation biomarkers identified from 121 studies, 37 single gene and seven multiple gene biomarkers were significantly associated with CRC prognosis after adjustment for clinical variables. Moreover, the directions of these associations with prognosis remained consistent between the original studies and our validation analyses. Seven single biomarkers and two multi-biomarker signatures were significantly associated with CRC prognosis in the meta-analysis, with a relatively strong level of evidence for CDKN2A, WNT5A, MLH1, and EVL. INTERPRETATION: In a comprehensive evaluation of the so far identified gene methylation biomarkers for CRC prognosis, we identified candidates with potential clinical relevance for further investigation. FUNDING: The German Research Council, the Interdisciplinary Research Program of the National Center for Tumor Diseases (NCT), Germany, the German Federal Ministry of Education and Research.

Regression-based Deep-Learning predicts molecular biomarkers from pathology slides.

Deep Learning (DL) can predict biomarkers from cancer histopathology. Several clinically approved applications use this technology. Most approaches, however, predict categorical labels, whereas biomarkers are often continuous measurements. We hypothesize that regression-based DL outperforms classification-based DL. Therefore, we develop and evaluate a self-supervised attention-based weakly supervised regression method that predicts continuous biomarkers directly from 11,671 images of patients across nine cancer types. We test our method for multiple clinically and biologically relevant biomarkers: homologous recombination deficiency score, a clinically used pan-cancer biomarker, as well as markers of key biological processes in the tumor microenvironment. Using regression significantly enhances the accuracy of biomarker prediction, while also improving the predictions' correspondence to regions of known clinical relevance over classification. In a large cohort of colorectal cancer patients, regression-based prediction scores provide a higher prognostic value than classification-based scores. Our open-source regression approach offers a promising alternative for continuous biomarker analysis in computational pathology.

End-to-end prognostication in colorectal cancer by deep learning: a retrospective, multicentre study.

BACKGROUND: Precise prognosis prediction in patients with colorectal cancer (ie, forecasting survival) is pivotal for individualised treatment and care. Histopathological tissue slides of colorectal cancer specimens contain rich prognostically relevant information. However, existing studies do not have multicentre external validation with real-world sample processing protocols, and algorithms are not yet widely used in clinical routine. METHODS: In this retrospective, multicentre study, we collected tissue samples from four groups of patients with resected colorectal cancer from Australia, Germany, and the USA. We developed and externally validated a deep learning-based prognostic-stratification system for automatic prediction of overall and cancer-specific survival in patients with resected colorectal cancer. We used the model-predicted risk scores to stratify patients into different risk groups and compared survival outcomes between these groups. Additionally, we evaluated the prognostic value of these risk groups after adjusting for established prognostic variables. FINDINGS: We trained and validated our model on a total of 4428 patients. We found that patients could be divided into high-risk and low-risk groups on the basis of the deep learning-based risk score. On the internal test set, the group with a high-risk score had a worse prognosis than the group with a low-risk score, as reflected by a hazard ratio (HR) of 4·50 (95% CI 3·33-6·09) for overall survival and 8·35 (5·06-13·78) for disease-specific survival (DSS). We found consistent performance across three large external test sets. In a test set of 1395 patients, the high-risk group had a lower DSS than the low-risk group, with an HR of 3·08 (2·44-3·89). In two additional test sets, the HRs for DSS were 2·23 (1·23-4·04) and 3·07 (1·78-5·3). We showed that the prognostic value of the deep learning-based risk score is independent of established clinical risk factors. INTERPRETATION: Our findings indicate that attention-based self-supervised deep learning can robustly offer a prognosis on clinical outcomes in patients with colorectal cancer, generalising across different populations and serving as a potentially new prognostic tool in clinical decision making for colorectal cancer management. We release all source codes and trained models under an open-source licence, allowing other researchers to reuse and build upon our work. FUNDING: The German Federal Ministry of Health, the Max-Eder-Programme of German Cancer Aid, the German Federal Ministry of Education and Research, the German Academic Exchange Service, and the EU.

Deep learning for dual detection of microsatellite instability and POLE mutations in colorectal cancer histopathology.

In the spectrum of colorectal tumors, microsatellite-stable (MSS) tumors with DNA polymerase ε (POLE) mutations exhibit a hypermutated profile, holding the potential to respond to immunotherapy similarly to their microsatellite-instable (MSI) counterparts. Yet, due to their rarity and the associated testing costs, systematic screening for these mutations is not commonly pursued. Notably, the histopathological phenotype resulting from POLE mutations is theorized to resemble that of MSI. This resemblance not only could facilitate their detection by a transformer-based Deep Learning (DL) system trained on MSI pathology slides, but also indicates the possibility for MSS patients with POLE mutations to access enhanced treatment options, which might otherwise be overlooked. To harness this potential, we trained a Deep Learning classifier on a large dataset with the ground truth for microsatellite status and subsequently validated its capabilities for MSI and POLE detection across three external cohorts. Our model accurately identified MSI status in both the internal and external resection cohorts using pathology images alone. Notably, with a classification threshold of 0.5, over 75% of POLE driver mutant patients in the external resection cohorts were flagged as "positive" by a DL system trained on MSI status. In a clinical setting, deploying this DL model as a preliminary screening tool could facilitate the efficient identification of clinically relevant MSI and POLE mutations in colorectal tumors, in one go.

Encrypted federated learning for secure decentralized collaboration in cancer image analysis.

Artificial intelligence (AI) has a multitude of applications in cancer research and oncology. However, the training of AI systems is impeded by the limited availability of large datasets due to data protection requirements and other regulatory obstacles. Federated and swarm learning represent possible solutions to this problem by collaboratively training AI models while avoiding data transfer. However, in these decentralized methods, weight updates are still transferred to the aggregation server for merging the models. This leaves the possibility for a breach of data privacy, for example by model inversion or membership inference attacks by untrusted servers. Somewhat-homomorphically-encrypted federated learning (SHEFL) is a solution to this problem because only encrypted weights are transferred, and model updates are performed in the encrypted space. Here, we demonstrate the first successful implementation of SHEFL in a range of clinically relevant tasks in cancer image analysis on multicentric datasets in radiology and histopathology. We show that SHEFL enables the training of AI models which outperform locally trained models and perform on par with models which are centrally trained. In the future, SHEFL can enable multiple institutions to co-train AI models without forsaking data governance and without ever transmitting any decryptable data to untrusted servers.