Three artificial intelligence data challenges based on CT and MRI

Advancements in the diagnosis of biliopancreatic diseases: A comparative review and study on future insights

Owing to the complex and often asymptomatic presentations, the diagnosis of biliopancreatic diseases, including pancreatic and biliary malignancies, remains challenging. Recent technological advancements have remarkably improved the diagnostic accuracy and patient outcomes in these diseases. This review explores key advancements in diagnostic modalities, including biomarkers, imaging techniques, and artificial intelligence (AI)-based technologies. Biomarkers, such as cancer antigen 19-9, KRAS mutations, and inflammatory markers, provide crucial insights into disease progression and treatment responses. Advanced imaging modalities include enhanced computed tomography (CT), positron emission tomography-CT, magnetic resonance cholangiopancreatography, and endoscopic ultrasound. AI integration in imaging and pathology has enhanced diagnostic precision through deep learning algorithms that analyze medical images, automate routine diagnostic tasks, and provide predictive analytics for personalized treatment strategies. The applications of these technologies are diverse, ranging from early cancer detection to therapeutic guidance and real-time imaging. Biomarker-based liquid biopsies and AI-assisted imaging tools are essential for non-invasive diagnostics and individualized patient management. Furthermore, AI-driven models are transforming disease stratification, thus enhancing risk assessment and decision-making. Future studies should explore standardizing biomarker validation, improving AI-driven diagnostics, and expanding the accessibility of advanced imaging technologies in resource-limited settings. The continued development of non-invasive diagnostic techniques and precision medicine approaches is crucial for optimizing the detection and management of biliopancreatic diseases. Collaborative efforts between clinicians, researchers, and industry stakeholders will be pivotal in applying these advancements in clinical practice.

Artificial intelligence in musculoskeletal applications: a primer for radiologists

As an umbrella term, artificial intelligence (AI) covers machine learning and deep learning. This review aimed to elaborate on these terms to act as a primer for radiologists to learn more about the algorithms commonly used in musculoskeletal radiology. It also aimed to familiarize them with the common practices and issues in the use of AI in this domain.

Detection and characterization of pancreatic lesion with artificial intelligence: The SFR 2023 artificial intelligence data challenge

PURPOSE

The purpose of the 2023 SFR data challenge was to invite researchers to develop artificial intelligence (AI) models to identify the presence of a pancreatic mass and distinguish between benign and malignant pancreatic masses on abdominal computed tomography (CT) examinations.

MATERIALS AND METHODS

Anonymized abdominal CT examinations acquired during the portal venous phase were collected from 18 French centers. Abdominal CT examinations were divided into three groups including CT examinations with no lesion, CT examinations with benign pancreatic mass, or CT examinations with malignant pancreatic mass. Each team included at least one radiologist, one data scientist, and one engineer. Pancreatic lesions were annotated by expert radiologists. CT examinations were distributed in balanced batches via a Health Data Hosting certified platform. Data were distributed into four batches, two for training, one for internal evaluation, and one for the external evaluation. Training used 83 % of the data from 14 centers and external evaluation used data from the other four centers. The metric (i.e., final score) used to rank the participants was a weighted average of mean sensitivity, mean precision and mean area under the curve.

RESULTS

A total of 1037 abdominal CT examinations were divided into two training sets (including 500 and 232 CT examinations), an internal evaluation set (including 139 CT examinations), and an external evaluation set (including 166 CT examinations). The training sets were distributed on September 7 and October 13, 2023, and evaluation sets on October 15, 2023. Ten teams with a total of 93 members participated to the data challenge, with the best final score being 0.72.

CONCLUSION

This SFR 2023 data challenge based on multicenter CT data suggests that the use of AI for pancreatic lesions detection is possible on real data, but the distinction between benign and malignant pancreatic lesions remains challenging.

The STOIC2021 COVID-19 AI challenge: Applying reusable training methodologies to private data

Challenges drive the state-of-the-art of automated medical image analysis. The quantity of public training data that they provide can limit the performance of their solutions. Public access to the training methodology for these solutions remains absent. This study implements the Type Three (T3) challenge format, which allows for training solutions on private data and guarantees reusable training methodologies. With T3, challenge organizers train a codebase provided by the participants on sequestered training data. T3 was implemented in the STOIC2021 challenge, with the goal of predicting from a computed tomography (CT) scan whether subjects had a severe COVID-19 infection, defined as intubation or death within one month. STOIC2021 consisted of a Qualification phase, where participants developed challenge solutions using 2000 publicly available CT scans, and a Final phase, where participants submitted their training methodologies with which solutions were trained on CT scans of 9724 subjects. The organizers successfully trained six of the eight Final phase submissions. The submitted codebases for training and running inference were released publicly. The winning solution obtained an area under the receiver operating characteristic curve for discerning between severe and non-severe COVID-19 of 0.815. The Final phase solutions of all finalists improved upon their Qualification phase solutions.

Dynamic radiological features predict pathological response after neoadjuvant immunochemotherapy in esophageal squamous cell carcinoma

BACKGROUND

Neoadjuvant immunochemotherapy (NICT) plus esophagectomy has emerged as a promising treatment option for locally advanced esophageal squamous cell carcinoma (LA-ESCC). Pathologic complete response (pCR) is a key indicator associated with great efficacy and overall survival (OS). However, there are insufficient indicators for the reliable assessment of pCR.

METHODS

192 patients with LA-ESCC treated with NICT from December 2019 to October 2023 were recruited. According to pCR status, patients were categorized into pCR group (22.92%) and non-pCR group (77.08%). Radiological features of pretreatment and preoperative CT images were extracted. Logistic and COX regressions were trained to predict pathological response and prognosis, respectively.

RESULTS

Four of the selected radiological features were combined to construct an ESCC preoperative imaging score (ECPI-Score). Logistic models revealed independent associations of ECPI-Score and vascular sign with pCR, with AUC of 0.918 in the training set and 0.862 in the validation set, respectively. After grouping by ECPI-Score, a higher proportion of pCR was observed among the high-ECPI group and negative vascular sign. Kaplan Meier analysis demonstrated that recurrence-free survival (RFS) with negative vascular sign was significantly better than those with positive (P = 0.038), but not for OS (P = 0.310).

CONCLUSIONS

This study demonstrates dynamic radiological features are independent predictors of pCR for LA-ESCC treated with NICT. It will guide clinicians to make accurate treatment plans.

Enhanced CT imaging artificial neural network coronary artery calcification score assisted diagnosis

BACKGROUND

The study of coronary artery calcification (CAC) may assist in identifying additional coronary artery problem protective factors. On the contrary side, due to the wide variety of CAC as individuals, CAC research is difficult. Due to this, evaluating data for investigation is becoming complicated.

OBJECTIVE

To use a multi-layer perceptron, we investigated the accuracy and reliability of synthetic CAC coursework or hazard classification in pre or alors chest computerized tomography (CT) of arrangements resolutions in this analysis.

method

Photographs of the chest from similar individuals as well as calcium-just and non-gated pictures were incorporated. This cut thickness ordered CT pictures (bunch A: 1 mm; bunch B: 3 mm). The CAC rating was determined utilizing calcification score picture information, and became standard for tests. While the control treatment's machine learning program was created using 170 computed tomography pictures and evaluated using 144 scans, group A's machine learning algorithm was created using 150 chest CT diagnostic tests.

RESULTS

334 external related pictures (100 μm: 117; 0.5 mm x: 117) of 117 individuals and 612 inside design organizing (1 mm: 294; mm3: 314) of 406 patients were surveyed. Pack B had 0.94, however, tests An and b had 0.90 (95% CI: 0.85-0.93) ICCs between significant learning and gold expenses (0.92-0.96). Dull Altman plots agreed well. A machine teaching approach successfully identified 71% of cases in category A is 81% of patients in section B again for cardiac risk class.

CONCLUSION

Regression risk evaluation algorithms could assist in categorizing cardiorespiratory individuals into distinct risk groups and conveniently personalize the treatments to the patient's circumstances. The models would be based on information gathered through CAC. On both 1 and 3-mm scanners, the automatic determination of a CAC value and cardiovascular events categorization that used a depth teaching approach was reliable and precise. The layer thickness of 0.5 mm on chest CT was slightly less accurate in CAC detection and risk evaluation.

Detection and quantification of pulmonary embolism with artificial intelligence: The SFR 2022 artificial intelligence data challenge

PURPOSE

In 2022, the French Society of Radiology together with the French Society of Thoracic Imaging and CentraleSupelec organized their 13th data challenge. The aim was to aid in the diagnosis of pulmonary embolism, by identifying the presence of pulmonary embolism and by estimating the ratio between right and left ventricular (RV/LV) diameters, and an arterial obstruction index (Qanadli's score) using artificial intelligence.

MATERIALS AND METHODS

The data challenge was composed of three tasks: the detection of pulmonary embolism, the RV/LV diameter ratio, and Qanadli's score. Sixteen centers all over France participated in the inclusion of the cases. A health data hosting certified web platform was established to facilitate the inclusion process of the anonymized CT examinations in compliance with general data protection regulation. CT pulmonary angiography images were collected. Each center provided the CT examinations with their annotations. A randomization process was established to pool the scans from different centers. Each team was required to have at least a radiologist, a data scientist, and an engineer. Data were provided in three batches to the teams, two for training and one for evaluation. The evaluation of the results was determined to rank the participants on the three tasks.

RESULTS

A total of 1268 CT examinations were collected from the 16 centers following the inclusion criteria. The dataset was split into three batches of 310, 580 and 378 C T examinations provided to the participants respectively on September 5, 2022, October 7, 2022 and October 9, 2022. Seventy percent of the data from each center were used for training, and 30% for the evaluation. Seven teams with a total of 48 participants including data scientists, researchers, radiologists and engineering students were registered for participation. The metrics chosen for evaluation included areas under receiver operating characteristic curves, specificity and sensitivity for the classification task, and the coefficient of determination r2 for the regression tasks. The winning team achieved an overall score of 0.784.

CONCLUSION

This multicenter study suggests that the use of artificial intelligence for the diagnosis of pulmonary embolism is possible on real data. Moreover, providing quantitative measures is mandatory for the interpretability of the results, and is of great aid to the radiologists especially in emergency settings.

Generative adversarial networks (GAN)-based data augmentation of rare liver cancers: The SFR 2021 Artificial Intelligence Data Challenge

PURPOSE

The 2021 edition of the Artificial Intelligence Data Challenge was organized by the French Society of Radiology together with the Centre National d'Études Spatiales and CentraleSupélec with the aim to implement generative adversarial networks (GANs) techniques to provide 1000 magnetic resonance imaging (MRI) cases of macrotrabecular-massive (MTM) hepatocellular carcinoma (HCC), a rare and aggressive subtype of HCC, generated from a limited number of real cases from multiple French centers.

MATERIALS AND METHODS

A dedicated platform was used by the seven inclusion centers to securely upload their anonymized MRI examinations including all three cross-sectional images (one late arterial and one portal-venous phase T1-weighted images and one fat-saturated T2-weighted image) in compliance with general data protection regulation. The quality of the database was checked by experts and manual delineation of the lesions was performed by the expert radiologists involved in each center. Multidisciplinary teams competed between October 11^th, 2021 and February 13^th, 2022.

RESULTS

A total of 91 MTM-HCC datasets of three images each were collected from seven French academic centers. Six teams with a total of 28 individuals participated in this challenge. Each participating team was asked to generate one thousand 3-image cases. The qualitative evaluation was performed by three radiologists using the Likert scale on ten randomly selected cases generated by each participant. A quantitative evaluation was also performed using two metrics, the Frechet inception distance and a leave-one-out accuracy of a 1-Nearest Neighbor algorithm.

CONCLUSION

This data challenge demonstrates the ability of GANs techniques to generate a large number of images from a small sample of imaging examinations of a rare malignant tumor.

Artificial intelligence in lung cancer: current applications and perspectives

Artificial intelligence (AI) has been a very active research topic over the last years and thoracic imaging has particularly benefited from the development of AI and in particular deep learning. We have now entered a phase of adopting AI into clinical practice. The objective of this article was to review the current applications and perspectives of AI in thoracic oncology. For pulmonary nodule detection, computer-aided detection (CADe) tools have been commercially available since the early 2000s. The more recent rise of deep learning and the availability of large annotated lung nodule datasets have allowed the development of new CADe tools with fewer false-positive results per examination. Classical machine learning and deep-learning methods were also used for pulmonary nodule segmentation allowing nodule volumetry and pulmonary nodule characterization. For pulmonary nodule characterization, radiomics and deep-learning approaches were used. Data from the National Lung Cancer Screening Trial (NLST) allowed the development of several computer-aided diagnostic (CADx) tools for diagnosing lung cancer on chest computed tomography. Finally, AI has been used as a means to perform virtual biopsies and to predict response to treatment or survival. Thus, many detection, characterization and stratification tools have been proposed, some of which are commercially available.

Artificial intelligence in musculoskeletal oncology imaging: A critical review of current applications

Artificial intelligence (AI) is increasingly being studied in musculoskeletal oncology imaging. AI has been applied to both primary and secondary bone tumors and assessed for various predictive tasks that include detection, segmentation, classification, and prognosis. Still, in the field of clinical research, further efforts are needed to improve AI reproducibility and reach an acceptable level of evidence in musculoskeletal oncology. This review describes the basic principles of the most common AI techniques, including machine learning, deep learning and radiomics. Then, recent developments and current results of AI in the field of musculoskeletal oncology are presented. Finally, limitations and future perspectives of AI in this field are discussed.

Artificial intelligence in adrenal imaging: A critical review of current applications

In the elective field of adrenal imaging, artificial intelligence (AI) can be used for adrenal lesion detection, characterization, hypersecreting syndrome management and patient follow-up. Although a perfect AI tool that includes all required steps from detection to analysis does not exist yet, multiple AI algorithms have been developed and tested with encouraging results. However, AI in this setting is still at an early stage. In this regard, most published studies about AI in adrenal gland imaging report preliminary results that do not have yet daily applications in clinical practice. In this review, recent developments and current results of AI in the field of adrenal imaging are presented. Limitations and future perspectives of AI are discussed.

Artificial Intelligence in Dermatology Image Analysis: Current Developments and Future Trends

BACKGROUND

Thanks to the rapid development of computer-based systems and deep-learning-based algorithms, artificial intelligence (AI) has long been integrated into the healthcare field. AI is also particularly helpful in image recognition, surgical assistance and basic research. Due to the unique nature of dermatology, AI-aided dermatological diagnosis based on image recognition has become a modern focus and future trend. Key scientific concepts of review: The use of 3D imaging systems allows clinicians to screen and label skin pigmented lesions and distributed disorders, which can provide an objective assessment and image documentation of lesion sites. Dermatoscopes combined with intelligent software help the dermatologist to easily correlate each close-up image with the corresponding marked lesion in the 3D body map. In addition, AI in the field of prosthetics can assist in the rehabilitation of patients and help to restore limb function after amputation in patients with skin tumors.

THE AIM OF THE STUDY

For the benefit of patients, dermatologists have an obligation to explore the opportunities, risks and limitations of AI applications. This study focuses on the application of emerging AI in dermatology to aid clinical diagnosis and treatment, analyzes the current state of the field and summarizes its future trends and prospects so as to help dermatologists realize the impact of new technological innovations on traditional practices so that they can embrace and use AI-based medical approaches more quickly.

Can AI predict epithelial lesion categories via automated analysis of cervical biopsies: The TissueNet challenge?

The French Society of Pathology (SFP) organized its first data challenge in 2020 with the help of the Health Data Hub (HDH). The organization of this event first consisted of recruiting nearly 5000 cervical biopsy slides obtained from 20 pathology centers. After ensuring that patients did not refuse to include their slides in the project, the slides were anonymized, digitized, and annotated by expert pathologists, and finally uploaded to a data challenge platform for competitors from around the world. Competing teams had to develop algorithms that could distinguish 4 diagnostic classes in cervical epithelial lesions. Among the many submissions from competitors, the best algorithms achieved an overall score close to 95%. The final part of the competition lasted only 6 weeks, and the goal of SFP and HDH is now to allow for the collection to be published in open access for the scientific community. In this report, we have performed a "post-competition analysis" of the results. We first described the algorithmic pipelines of 3 top competitors. We then analyzed several difficult cases that even the top competitors could not predict correctly. A medical committee of several expert pathologists looked for possible explanations for these erroneous results by reviewing the images, and we present their findings here targeted for a large audience of pathologists and data scientists in the field of digital pathology.

Validation of a deep learning segmentation algorithm to quantify the skeletal muscle index and sarcopenia in metastatic renal carcinoma

OBJECTIVES

To validate a deep learning (DL) algorithm for measurement of skeletal muscular index (SMI) and prediction of overall survival in oncology populations.

METHODS

A retrospective single-center observational study included patients with metastatic renal cell carcinoma between 2007 and 2019. A set of 37 patients was used for technical validation of the algorithm, comparing manual vs DL-based evaluations. Segmentations were compared using mean Dice similarity coefficient (DSC), SMI using concordance correlation coefficient (CCC) and Bland-Altman plots. Overall survivals (OS) were compared using log-rank (Kaplan-Meier) and Mann-Whitney tests. Generalizability of the prognostic value was tested in an independent validation population (N = 87).

RESULTS

Differences between two manual segmentations (DSC = 0.91, CCC = 0.98 for areas) or manual vs. automated segmentation (DSC = 0.90, CCC = 0.98 for areas, CCC = 0.97 for SMI) had the same order of magnitude. Bland-Altman plots showed a mean difference of -3.33 cm² [95%CI: -15.98, 9.1] between two manual segmentations, and -3.28 cm² [95% CI: -14.77, 8.21] for manual vs. automated segmentations. With each method, 20/37 (56%) patients were classified as sarcopenic. Sarcopenic vs. non-sarcopenic groups had statistically different survival curves with median OS of 6.0 vs. 12.5 (p = 0.008) and 6.0 vs. 13.9 (p = 0.014) months respectively for manual and DL methods. In the independent validation population, sarcopenic patients according to DL had a lower OS (10.7 vs. 17.3 months, p = 0.033).

CONCLUSION

A DL algorithm allowed accurate estimation of SMI compared to manual reference standard. The DL-calculated SMI demonstrated a prognostic value in terms of OS.

KEY POINTS

• A deep learning algorithm allows accurate estimation of skeletal muscle index compared to a manual reference standard with a concordance correlation coefficient of 0.97. • Sarcopenic patients according to SMI thresholds after segmentation by the deep learning algorithm had statistically significantly lower overall survival compared to non-sarcopenic patients.

External validation of a commercially available deep learning algorithm for fracture detection in children: Fracture detection with a deep learning algorithm

PURPOSE

The purpose of this study was to conduct an external validation of a fracture assessment deep learning algorithm (Rayvolve®) using digital radiographs from a real-life cohort of children presenting routinely to the emergency room.

MATERIALS AND METHODS

This retrospective study was conducted on 2634 radiography sets (5865 images) from 2549 children (1459 boys, 1090 girls; mean age, 8.5 ± 4.5 [SD] years; age range: 0-17 years) referred by the pediatric emergency room for trauma. For each set was recorded whether one or more fractures were found, the number of fractures, and their location found by the senior radiologists and the algorithm. Using the senior radiologist diagnosis as the standard of reference, the diagnostic performance of deep learning algorithm (Rayvolve®) was calculated via three different approaches: a detection approach (presence/absence of a fracture as a binary variable), an enumeration approach (exact number of fractures detected) and a localization approach (focusing on whether the detected fractures were correctly localized). Subgroup analyses were performed according to the presence of a cast or not, age category (0-4 vs. 5-18 years) and anatomical region.

RESULTS

Regarding detection approach, the deep learning algorithm yielded 95.7% sensitivity (95% CI: 94.0-96.9), 91.2% specificity (95% CI: 89.8-92.5) and 92.6% accuracy (95% CI: 91.5-93.6). Regarding enumeration and localization approaches, the deep learning algorithm yielded 94.1% sensitivity (95% CI: 92.1-95.6), 88.8% specificity (95% CI: 87.3-90.2) and 90.4% accuracy (95% CI: 89.2-91.5) for both approaches. Regarding age-related subgroup analyses, the deep learning algorithm yielded greater sensitivity and negative predictive value in the 5-18-years age group than in the 0-4-years age group for the detection approach (P < 0.001 and P = 0.002) and for the enumeration and localization approaches (P = 0.012 and P = 0.028). The high negative predictive value was robust, persisting in all of the subgroup analyses, except for patients with casts (P = 0.001 for the detection approach and P < 0.001 for the enumeration and localization approaches).

CONCLUSION

The Rayvolve® deep learning algorithm is very reliable for detecting fractures in children, especially in those older than 4 years and without cast.

Breast nodule classification with two-dimensional ultrasound using Mask-RCNN ensemble aggregation

PURPOSE

The purpose of this study was to create a deep learning algorithm to infer the benign or malignant nature of breast nodules using two-dimensional B-mode ultrasound data initially marked as BI-RADS 3 and 4.

MATERIALS AND METHODS

An ensemble of mask region-based convolutional neural networks (Mask-RCNN) combining nodule segmentation and classification were trained to explicitly localize the nodule and generate a probability of the nodule to be malignant on two-dimensional B-mode ultrasound. These probabilities were aggregated at test time to produce final results. Resulting inferences were assessed using area under the curve (AUC).

RESULTS

A total of 460 ultrasound images of breast nodules classified as BI-RADS 3 or 4 were included. There were 295 benign and 165 malignant breast nodules used for training and validation, and another 137 breast nodules images used for testing. As a part of the challenge, the distribution of benign and malignant breast nodules in the test database remained unknown. The obtained AUC was 0.69 (95% CI: 0.57-0.82) on the training set and 0.67 on the test set.

CONCLUSION

The proposed deep learning solution helps classify benign and malignant breast nodules based solely on two-dimensional ultrasound images initially marked as BIRADS 3 and 4.

Three artificial intelligence data challenges based on CT and ultrasound

PURPOSE

The 2020 edition of these Data Challenges was organized by the French Society of Radiology (SFR), from September 28 to September 30, 2020. The goals were to propose innovative artificial intelligence solutions for the current relevant problems in radiology and to build a large database of multimodal medical images of ultrasound and computed tomography (CT) on these subjects from several French radiology centers.

MATERIALS AND METHODS

This year the attempt was to create data challenge objectives in line with the clinical routine of radiologists, with less preprocessing of data and annotation, leaving a large part of the preprocessing task to the participating teams. The objectives were proposed by the different organizations depending on their core areas of expertise. A dedicated platform was used to upload the medical image data, to automatically anonymize the uploaded data.

RESULTS

Three challenges were proposed including classification of benign or malignant breast nodules on ultrasound examinations, detection and contouring of pathological neck lymph nodes from cervical CT examinations and classification of calcium score on coronary calcifications from thoracic CT examinations. A total of 2076 medical examinations were included in the database for the three challenges, in three months, by 18 different centers, of which 12% were excluded. The 39 participants were divided into six multidisciplinary teams among which the coronary calcification score challenge was solved with a concordance index > 95%, and the other two with scores of 67% (breast nodule classification) and 63% (neck lymph node calcifications).

Automatic coronary artery calcium scoring from unenhanced-ECG-gated CT using deep learning

PURPOSE

The purpose of this study was to develop and evaluate an algorithm that can automatically estimate the amount of coronary artery calcium (CAC) from unenhanced electrocardiography (ECG)-gated computed tomography (CT) cardiac volume acquisitions by using convolutional neural networks (CNN).

MATERIALS AND METHODS

The method used a set of five CNN with three-dimensional (3D) U-Net architecture trained on a database of 783 CT examinations to detect and segment coronary artery calcifications in a 3D volume. The Agatston score, the conventional CAC scoring, was then computed slice by slice from the resulting segmentation mask and compared to the ground truth manually estimated by radiologists. The quality of the estimation was assessed with the concordance index (C-index) on CAC risk category on a separate testing set of 98 independent CT examinations.

RESULTS

The final model yielded a C-index of 0.951 on the testing set. The remaining errors of the method were mainly observed on small-size and/or low-density calcifications, or calcifications located near the mitral valve or ring.

CONCLUSION

The deep learning-based method proposed here to compute automatically the CAC score from unenhanced-ECG-gated cardiac CT is fast, robust and yields accuracy similar to those of other artificial intelligence methods, which could improve workflow efficiency, eliminating the time spent on manually selecting coronary calcifications to compute the Agatston score.

Artificial intelligence: a critical review of current applications in pancreatic imaging

The applications of artificial intelligence (AI), including machine learning and deep learning, in the field of pancreatic disease imaging are rapidly expanding. AI can be used for the detection of pancreatic ductal adenocarcinoma and other pancreatic tumors but also for pancreatic lesion characterization. In this review, the basic of radiomics, recent developments and current results of AI in the field of pancreatic tumors are presented. Limitations and future perspectives of AI are discussed.

Automatic cervical lymphadenopathy segmentation from CT data using deep learning

PURPOSE

The purpose of this study was to develop a fast and automatic algorithm to detect and segment lymphadenopathy from head and neck computed tomography (CT) examination.

MATERIALS AND METHODS

An ensemble of three convolutional neural networks (CNNs) based on a U-Net architecture were trained to segment the lymphadenopathies in a fully supervised framework. The resulting predictions were assessed using the Dice similarity coefficient (DSC) on examinations presenting one or more adenopathies. On examinations without adenopathies, the score was given by the formula M/(M+A) where M was the mean adenopathy volume per patient and A the volume segmented by the algorithm. The networks were trained on 117 annotated CT acquisitions.

RESULTS

The test set included 150 additional CT acquisitions unseen during the training. The performance on the test set yielded a mean score of 0.63.

CONCLUSION

Despite limited available data and partial annotations, our CNN based approach achieved promising results in the task of cervical lymphadenopathy segmentation. It has the potential to bring precise quantification to the clinical workflow and to assist the clinician in the detection task.

Artificial intelligence solution to classify pulmonary nodules on CT

PURPOSE

The purpose of this study was to create an algorithm to detect and classify pulmonary nodules in two categories based on their volume greater than 100 mm³ or not, using machine learning and deep learning techniques.

MATERIALS AND METHOD

The dataset used to train the model was provided by the organization team of the SFR (French Radiological Society) Data Challenge 2019. An asynchronous and parallel 3-stages pipeline was developed to process all the data (a data "pre-processing" stage; a "nodule detection" stage; a "classifier" stage). Lung segmentation was achieved using 3D U-NET algorithm; nodule detection was done using 3D Retina-UNET and classifier stage with a support vector machine algorithm on selected features. Performances were assessed using area under receiver operating characteristics curve (AUROC).

RESULTS

The pipeline showed good performance for pathological nodule detection and patient diagnosis. With the preparation dataset, an AUROC of 0.9058 (95% confidence interval [CI]: 0.8746-0.9362) was obtained, 87% yielding accuracy (95% CI: 84.83%-91.03%) for the "nodule detection" stage, corresponding to 86% specificity (95% CI: 82%-92%) and 89% sensitivity (95% CI: 84.83%-91.03%).

CONCLUSION

A fully functional pipeline using 3D U-NET, 3D Retina-UNET and classifier stage with a support vector machine algorithm was developed, resulting in high capabilities for pulmonary nodule classification.