Relative rates of missed diagnosis for colonoscopy, barium enema, and flexible sigmoidoscopy in 379 patients with colorectal cancer

Development of an ensemble CNN model with explainable AI for the classification of gastrointestinal cancer

The implementation of AI assisted cancer detection systems in clinical environments has faced numerous hurdles, mainly because of the restricted explainability of their elemental mechanisms, even though such detection systems have proven to be highly effective. Medical practitioners are skeptical about adopting AI assisted diagnoses as due to the latter's inability to be transparent about decision making processes. In this respect, explainable artificial intelligence (XAI) has emerged to provide explanations for model predictions, thereby overcoming the computational black box problem associated with AI systems. In this particular research, the focal point has been the exploration of the Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) approaches which enable model prediction explanations. This study used an ensemble model consisting of three convolutional neural networks(CNN): InceptionV3, InceptionResNetV2 and VGG16, which was based on averaging techniques and by combining their respective predictions. These models were trained on the Kvasir dataset, which consists of pathological findings related to gastrointestinal cancer. An accuracy of 96.89% and F1-scores of 96.877% were attained by our ensemble model. Following the training of the ensemble model, we employed SHAP and LIME to analyze images from the three classes, aiming to provide explanations regarding the deterministic features influencing the model's predictions. The results obtained from this analysis demonstrated a positive and encouraging advancement in the exploration of XAI approaches, specifically in the context of gastrointestinal cancer detection within the healthcare domain.

Explainable AI-driven model for gastrointestinal cancer classification

Although the detection procedure has been shown to be highly effective, there are several obstacles to overcome in the usage of AI-assisted cancer cell detection in clinical settings. These issues stem mostly from the failure to identify the underlying processes. Because AI-assisted diagnosis does not offer a clear decision-making process, doctors are dubious about it. In this instance, the advent of Explainable Artificial Intelligence (XAI), which offers explanations for prediction models, solves the AI black box issue. The SHapley Additive exPlanations (SHAP) approach, which results in the interpretation of model predictions, is the main emphasis of this work. The intermediate layer in this study was a hybrid model made up of three Convolutional Neural Networks (CNNs) (InceptionV3, InceptionResNetV2, and VGG16) that combined their predictions. The KvasirV2 dataset, which comprises pathological symptoms associated to cancer, was used to train the model. Our combined model yielded an accuracy of 93.17% and an F1 score of 97%. After training the combined model, we use SHAP to analyze images from these three groups to provide an explanation of the decision that affects the model prediction.

Interval Colorectal Cancer 2006-2015: Novel Observations

BACKGROUND

Screening for colorectal cancer (CRC) with colonoscopy has been credited for the majority of the decline in mortality over the past 40 years. However, colonoscopy is hampered by frequent interval cancers (ICs) occurring within 36 months after the last examination. Risk factors for IC imply that most misses are amenable to improved colonoscopic techniques. This study had two primary objectives: (1) to determine whether the proportion of persons with ICs has declined in association with improved approaches to the quality of colonoscopy and (2) to determine unrecognized causes for missed diagnoses.

METHODS

This is a retrospective, single-center study of persons with CRC having been diagnosed within 6-36 months since the most recent colonoscopy between 2006 and 2015. Participants, ages 50 and 89 years, were evaluated specifically for technical features and findings of the most recent colonoscopy.

RESULTS

Thirty-nine persons met the inclusion criteria for IC. Mean age was 69.3 years, range 50-86 years, women were 58%, races included 95% Caucasian, 2.5% Asian, and 2.5% African-American, and history of a first-degree relative with CRC was 26%. Symptoms led to the interval diagnosis in 83%, including iron-deficient anemia (38%), bleeding (25%), abdominal/pelvic pain (18%), and change in bowel habit (2.5%). Preexisting neoplasia was present in 72%. The CRC was located in the proximal colon in 51%. Late-stage disease was present in 45%. When compared to persons with an initial diagnosis of screen-detected late-onset CRC, differences were noted for older age, more women, more with a family history of CRC, and more with late-stage disease. The ratio of interval cancers to new cancers ranged between 6 and 11.5%. Findings indicating a "difficult examination" were observed in 14 cases (36%) compared to "easy or no difficulty" examinations in 25 (65%) cases. In a biennial analysis of IC rates between 2006 and 2015, the percentage of IC remained stable between 6 and 11.5%.

CONCLUSIONS

ICs have been stable constituting 6-11.5% of all CRCs. A "difficult examination" was often associated with IC, may be a risk factor for this problem, and may be an important clue warranting further study.

Cell-surface markers for colon adenoma and adenocarcinoma

Early detection of colorectal cancer (CRC) is crucial for effective treatment. Among CRC screening techniques, optical colonoscopy is widely considered the gold standard. However, it is a costly and invasive procedure with a low rate of compliance. Our long-term goal is to develop molecular imaging agents for the non-invasive detection of CRC by molecular imaging-based colonoscopy using CT, MRI or fluorescence. To achieve this, cell surface targets must be identified and validated. Here, we report the discovery of cell-surface markers that distinguish CRC from surrounding tissues that could be used as molecular imaging targets. Profiling of mRNA expression microarray data from patient tissues including adenoma, adenocarcinoma, and normal gastrointestinal tissues was used to identify potential CRC specific cell-surface markers. Of the identified markers, six were selected for further validation (CLDN1, GPR56, GRM8, LY6G6D/F, SLCO1B3 and TLR4). Protein expression was confirmed by immunohistochemistry of patient tissues. Except for SLCO1B3, diffuse and low expression was observed for each marker in normal colon tissues. The three markers with the greatest protein overexpression were CLDN1, LY6G6D/F and TLR4, where at least one of these markers was overexpressed in 97% of the CRC samples. GPR56, LY6G6D/F and SLCO1B3 protein expression was significantly correlated with the proximal tumor location and with expression of mismatch repair genes. Marker expression was further validated in CRC cell lines. Hence, three cell-surface markers were discovered that distinguish CRC from surrounding normal tissues. These markers can be used to develop imaging or therapeutic agents targeted to the luminal surface of CRC.

Systematic analysis of missed colorectal cancer cases and common pitfalls in diagnosis

BACKGROUND

Missed colorectal cancer on endoscopic or radiological investigations may delay diagnosis and impact outcome. This study audits incidence of previous investigations in patients with colorectal cancer, considers outcome in 'missed' cancer cases and examines the diagnostic pathway in the derived case series to identify common pitfalls in diagnosis.

METHODS

Patients diagnosed with colorectal cancer in 2011 at a single National Health Service (NHS) Trust were reviewed. Incidence of endoscopic and radiological investigations in the 3 years preceding diagnosis and outcome data were collected. Cases of prior investigation not leading to diagnosis were considered 'missed' cancers and survival compared with 'detected' cases. The diagnostic pathway in each 'missed' case was reviewed.

RESULTS

395 colorectal cancer cases were studied. Eighteen (4.6%) patients underwent previous investigation including colonoscopy (n=4), flexible sigmoidoscopy (n=5), barium enema (n=5) and diagnostic abdominal CT scan (n=12), median 708 days prior to diagnosis. Previous investigation predicted reduced overall and disease-free survival (HR 2.07, p=0.04 and HR 2.66, p<0.0001), after age and gender adjustment. Ten different categories termed 'pitfalls' were derived from analysis of the diagnostic pathway. These included CT scanning for abdominal pain without further investigation (n=7), rectosigmoid cancer following a previous diagnosis of diverticular disease (n=4) and incomplete diagnostic investigations without adequate follow-up (n=3).

CONCLUSIONS

A proportion of patients diagnosed with colorectal cancer have previously been investigated for gastrointestinal symptoms and survival appears reduced in these patients. Regular audit and analysis of previous investigations can identify common pitfalls in diagnosis, which should be used to inform training and improve practice.