Differential diagnosis of Interstitial Lung Diseases using Deep Learning networks

Background. Abatacept has been suggested to be a safe and effective drug for the treatment of rheumatoid arthritis (RA) complicated by interstitial lung disease (ILD), one of the most severe extra-articular manifestations of RA. Objectives. In this prospective multicentre observational study, we aimed to evaluate the effectiveness and safety of abatacept on RA-ILD. Concurrently, the study aimed to evaluate the incidence of new diagnosis of ILD in RA patients treated with abatacept. Methods. We enrolled all consecutive RA patients already under treatment with abatacept for at least three months referred to 12 rheumatological centres in Italy. ILD patients were followed according to the current clinical practice: forced vital capacity (FVC%) was repeated every six months, while chest high resolution computed tomography (HRCT) was performed annually or when respiratory symptoms worsened. In patients without ILD at baseline, a careful assessment for respiratory symptoms and velcro crackles was performed at enrolment and every six months. HRCT was requested in case of new-onset dyspnoea, persistent dry cough, or detection of velcro crackles. Results. We enrolled 325 RA patients treated with abatacept. Among them, ILD was already present in 20.6% of cases. A usual interstitial pneumonia (UIP) pattern was detected 25.4% of cases. The two-year retention rate of abatacept was 79.6% ±3.8 and 76.8%±10.6 in patients without and with ILD, respectively, without difference between the two groups. Two cases of infectious pneumonia were reported in ILD group, requiring a transient discontinuation of abatacept. No patients discontinued abatacept for a worsening of lung function, and no episodes of acute exacerbation of ILD were observed during the study. At the end of follow-up, the radiological picture remained stable in 71.6% of subjects, improved in 13.4%, and worsened in 14.9%. A worsening at HRCT was recorded in 29.4% of cases of UIP, and only in 10% of patients with other ILD patterns (p not significant). Regarding lung function tests, FVC improved in 13.4% of cases, remained unchanged in 74.6%, and worsened in 11.9%. During the 24-month period of follow-up, 2 patients developed ILD, with an incidence of 0.39 new ILD cases/100 patients/year. Conclusions. Our study confirmed the safety and the effectiveness of abatacept in this population of difficult to treat RA patients. Of interest, ILD didn’t affect the retention rate of the drug. Moreover, lung function and HRCT features improved or remained stable in about 85% of patients, suggesting a role for abatacept in the treatment of RA-ILD. Concurrently, we observed a low rate of new cases of ILD, with an incidence rate of 0.39 new ILD cases/100 patients/year. Controlled studies, specifically developed in RA-ILD population, will be necessary to define the efficacy and safety of abatacept compared to other biologic DMARDs.

Early diagnosis of interstitial lung diseases secondary to connective tissue diseases is critical for the treatment and survival of patients. The symptoms, like dry cough and dyspnea, appear late in the clinical history and are not specific, moreover, the current approach to confirm the diagnosis of interstitial lung disease is based on high resolution computer tomography. However, computer tomography involves x-ray exposure for patients and high costs for the Health System, therefore preventing its use for a massive screening campaign in elder people.In this work we investigate the use of deep learning techniques for the classification of pulmonary sounds acquired from patients affected by connective tissue diseases. The novelty of the work consists of a suitably developed pre-processing pipeline for de-noising and data augmentation. The proposed approach is combined with a clinical study where the ground truth is represented by high resolution computer tomography. Various convolutional neural networks have provided an overall accuracy as high as 91% in the classification of lung sounds and have led to an overwhelming diagnostic accuracy in the range 91%−93%. Modern high performance hardware for edge computing can easily support our algorithms. This solution paves the way for a vast screening campaign of interstitial lung diseases in elder people on the basis of a non-invasive and cheap thoracic auscultation.

Early and accurate diagnosis of interstitial lung diseases (ILDs) is crucial for making treatment decisions, but can be challenging even for experienced radiologists. The diagnostic procedure is based on the detection and recognition of the different ILD pathologies in thoracic CT scans, yet their manifestation often appears similar. In this study, we propose the use of a deep purely convolutional neural network for the semantic segmentation of ILD patterns, as the basic component of a computer aided diagnosis system for ILDs. The proposed CNN, which consists of convolutional layers with dilated filters, takes as input a lung CT image of arbitrary size and outputs the corresponding label map. We trained and tested the network on a data set of 172 sparsely annotated CT scans, within a cross-validation scheme. The training was performed in an end-to-end and semisupervised fashion, utilizing both labeled and nonlabeled image regions. The experimental results show significant performance improvement with respect to the state of the art.

Current data have shown that lung ultrasound (LUS) is an emerging tool in the diagnosis of interstitial lung disease (ILD) by evaluating numbers of B-lines, pleural irregularities and nodules or consolidations. These features can distinguish between ILD and others different pulmonary conditions (i.e., patients with heart failure or end-stage renal disease accompanied by pulmonary congestion). Although there is not a standardized approach for diagnosis of ILD by LUS examination and most studies has been made on rheumatologic patients, we discuss utility of LUS in detecting ILD and its correlation to classical radiological finding, especially to high-resolution computed tomography (HRCT).

Interstitial lung disease (ILD) encompasses a heterogeneous group of over 200 diffuse parenchymal lung diseases. Accurate diagnosis of ILD is essential for appropriate treatment but can be challenging to achieve. For patients with suspected ILD, high-resolution computed tomography (HRCT) is a minimally invasive approach that allows for convenient repeat imaging. Compared to a conventional computed tomography (CT) protocol, the HRCT protocol for ILD patients involves two additional series with different technical requirements: a prone inspiration scan, and a supine expiration scan. The prone scan is useful when dependent opacification is observed on a supine chest CT scan. It facilitates the diagnosis of honeycombing, reducing observer variation in diagnosing idiopathic pulmonary fibrosis. The expiratory scan is useful for identifying air trapping, which is essential for the diagnosis of hypersensitivity pneumonitis. It is used to assess the three-density sign, which is specific for hypersensitivity pneumonitis and characterized by patchy distribution of normal-appearing lobules, ground glass opacities, and lobules with reduced lung density and vessel size. The HRCT ILD protocol should be performed regularly to assess changes in the extent of fibrosis and to diagnose lung cancer or acute exacerbation when clinically indicated.

Summary In this paper we describe the goals and the latest outcomes of the TA LISMAN project, the purpose of which is to furnish image-based diagnostic assistance for interstitial lung diseases (ILDs) with secondary data integration. Prototypes of the computer tools are implemented. High correct classification rates of lung tissue regions in high-resolution computed tomography (HRCT), based on a high-quality dataset built from clinical routine, suggests that computerised analysis of HRCT images with integration of the clinical context is ready to be used for computer-aided diagnosis of ILDs. As future tasks, implementation of multimodal retrieval of ILD cases and clinical evaluation of the software are planned to meet clinical needs.

Bronchoalveolar lavage (BAL) is a useful technique for differential diagnosis of various interstitial lung diseases (ILDs) and is usually realized by analysis of the differential cell count. This study was conducted to estimate the value of bronchoalveolar lavage fluid (BALF) total cell count (TCC) in the diagnosis of ILD. We analyzed 237 BAL samples from patients with ILD: sarcoidosis (SA), idiopathic pulmonary fibrosis (IPF), cryptogenic organizing pneumonia (COP), hypersensitivity pneumonitis (HP), chronic eosinophilic pneumonia (CEP), and smoking-related ILD (sr-ILD). The control group consisted of 30 healthy volunteers. The statistical analysis revealed significant differences in the BALF TCC between healthy controls and patients with SA, IPF, HP, COP, sr-ILD, and eosinophilic disorders (mean values 6.9 vs. 14.5, 22.5, 22.8, 20.7, 64.5, and 27.3 × 10(6), respectively). Logistic regression revealed a significant relation between the TCC and ILD diagnosis. We conclude that the TCC, as well as the value of total number of inflammatory cells, should be reported in the description of BAL.

Lungs play an important role in human respiratory system. There are diseases that affect the functioning of lungs. To analyse lung diseases in the chest region using X-ray based Computer-Aided Diagnosis (CAD) system, it is necessary to determine the lung regions subject to analysis. In this paper, an intelligent system is proposed for lung disease detection. In this paper, Interstitial Lung Disease (ILD) patterns are classified using Convolutional Neural Networks (CNN). The proposed system involves five convolutional layers and three dense layers. The performance of the classification demonstrates the potential of CNN in analysing lung patterns.

Introduction: High resolution computed tomography (HRCT) is essential in the diagnostic algorithm for idiopathic pulmonary fibrosis (IPF). If identifiable causes of interstitial lung disease (ILD) are absent, an HRCT showing a typical usual interstitial pneumonia (UIP) pattern is diagnostic for IPF. Here we report first experiences with a teleradiology-project that offers physicians the possibility for second opinion HRCT read regarding the presence of UIP pattern in patients with suspected IPF. Methods: HRCT were uploaded through a secured internet connection. Only images in original Digital Imaging and Communication in Medicine (DICOM) format were accepted. HRCTs were assessed by a radiologist and a pulmonologist, both experienced in the diagnosis of ILD. According to current guidelines each HRCT was rated showing either typical UIP pattern, possible UIP pattern, or being inconsistent with UIP pattern. Results: Between March 2015 and January 2017 we received 160 HRCTs for a second opinion read from physicians in 10 different Eastern European countries. All images were of diagnostic quality. In 58/160 (36%) patients a typical UIP pattern was diagnosed, possible UIP pattern was detected in 27/160 (17%), and in 75/160 (47%) patients the prevailing pattern was considered as inconsistent with UIP. Conclusions: Our data demonstrate a useful application of teleradiology to provide expert reading of HRCT in patients with suspected IPF, particularly when local radiological expertise is limited. This project is not intended to replace local multidisciplinary panels, but to support local panels by providing specific expertise. More experience will be needed to define the role of teleradiology in patients with suspected IPF.

Dasatinib is a second-generation tyrosine kinase inhibitor (TKI) used in chronic myelogenous leukemia (CML). While pleural effusion due to Dasatinib is well described in the literature, interstitial lung disease (ILD) caused by it is rare. A 60-year-old gentleman was on treatment with 100 mg of tablet Dasatinib per day for chronic myeloid leukemia. He presented to the outpatient department with history of progressive breathlessness over 2 months. High-resolution computerized tomography (HRCT) thorax revealed mild right-sided effusion and non-specific interstitial pneumonia (NSIP) pattern of ILD in the left lower lobe. Thoracocentesis of the right-sided pleural effusion showed exudative and lymphocytic rich pleural effusion. The effusion was negative for malignant cells or infection. Biopsy of the left lower lobe was consistent with the diagnosis of ILD. He was started on prednisolone which was gradually tapered and stopped. At 3 months, there was a complete resolution of the ILD and pleural effusion. Clinicians need to be aware about the pleuroparenchymal toxicities of Dasatinib. Early diagnosis and treatment with steroids can lead to complete resolution of the signs and symptoms.

In deep convolutional neural networks (DCNNs), model size and computation complexity are two important factors governing throughput and energy efficiency when deployed to hardware for inference. Recent works on compact DCNNs as well as pruning methods are effective, yet with drawbacks. For instance, more than half the size of all MobileNet models lies in their last two layers, mainly because compact separable convolution (CONV) layers are not applicable to their last fully connected (FC) layers. Also, in pruning methods, the compression is gained at the expense of irregularity in the DCNN architecture, which necessitates additional indexing memory to address nonzero weights, thereby increasing memory footprint, decompression delays, and energy consumption. In this article, we propose cyclic sparsely connected (CSC) architectures, with memory/computation complexity of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {O}(N\log {}N)$ </tex-math></inline-formula> , where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$N$ </tex-math></inline-formula> is the number of nodes/channels given a DCNN layer that, contrary to compact depthwise separable layers, can be used as an overlay for both FC and CONV layers of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {O}(N^{2})$ </tex-math></inline-formula> . Also, contrary to pruning methods, CSC architectures are structurally sparse and require no indexing due to their cyclic nature. We show that both standard convolution and depthwise convolution layers are special cases of the CSC layers, whose mathematical function, along with FC layers, can be unified into one single formulation and whose hardware implementation can be carried out under one arithmetic logic component. We examine the efficacy of the CSC architectures for compression of LeNet, AlexNet, and MobileNet DCNNs with precision ranging from 2 to 32 bits. More specifically, we surge upon the compact 8-bit quantized 0.5 MobileNet V1 and show that by compressing its last two layers with CSC architectures, the model is compressed by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\sim 1.5\times $ </tex-math></inline-formula> with a size of only 873 kB and little accuracy loss. Finally, we design a configurable hardware that implements all types of DCNN layers including FC, CONV, depthwise, CSC-FC, and CSC-CONV indistinguishably within a unified pipeline. We implement the hardware on a tiny Xilinx field-programmable gate array (FPGA) for total on-chip processing of the compressed MobileNet that, compared to the related work, has the highest Inference/J while utilizing the smallest FPGA.

Introduction: We reported a patient developed acute interstitial lung disease (ILD) after administration of Fluconazole for cryptococcal pneumonia on the basis of using Abemaciclib. The case underscored a possible concentration-dependent severe adverse reaction of Abemaciclib, and the importance of using drugs in combination with cautious. Case Description: A 75-year-old female developed high fever and dry cough after contact with pigeons, and admitted 1 week later on April, 2024. She received Abemaciclib for recurrence of breast cancer on the left chest wall from June, 2023. High-resolution computed tomography (HRCT) after administration revealed multiple solitary ground-glass nodules in both lungs. Further examinations identified negative autoantibodies, but a positive cryptococcal antigen. She was administrated with intravenous Fluconazole 400 milligram (mg) per day for the treatment of cryptococcal pneumonia. However, the patient continued with high fever, and developed dyspnea during the next two weeks. Her c-reactive protein (CRP) increased progressively (9→42mg/L), while series HRCT showed diffused ground-glass opacities and reticulars, especially in upper lobe of the left lung. Bronchoalveolar lavage (BAL) followed found elevated lymphocytes and neutrophils with a percentage of 15% and 54%, respectively. The metagenomic second generation sequencing of BAL detected 3 copies of Cryptococcus. Multidisciplinary discussion considered a diagnosis of acute interstitial lung disease (ILD), most likely caused by by adding Fluconazole on the basis of Abemaciclib. The patient discontinued Abemaciclib, and received Methylprednisolone 40mg per day. Her dyspena alleviated, with a normal body temperature and CRP. Methylprednisolone gradually tapered during the next 6-months follow-up, and the interstitial changes mostly resolved on October, 2024. Discussion: As an inhibitor of cyclin-dependent kinase, Abemaciclib can induce ILD with an incidence of 1.7%, after a median of 2-months administration, as observed by real-world cohorts. However, in this case, the patient did not manifest ILD during nearly 10-months treatment of Abemaciclib alone, but developed ILD acutely after receiving Fluconazole in combination. Fluconazole can increase the concentration of Abemaciclib by inhibiting cytochrome P450.This suggested that Abemaciclib might induce ILD in a concentration-dependent manner. Discontinuation of combination medication, and intervention with corticostroids were critical to reverse the poor prognosis. Figure1 Interstitial changes of the Apical segment. Compared to baseline HRCT (A), the Ground-glass opacities and reticulars developed rapidly after adding fluconazle on the basis of Abemaciclib (B and C), while improved after discontinuation of Abemaciclib and administration of Methylprednisolone (D).

To investigate the value of exhaled nitric oxide (eNO) in the diagnosis and differential diagnosis of interstitial lung disease (ILD). This study was conducted among 45 patients with interstitial lung disease, including 18 with connective tissue disease-related ILD (CTD-ILD) and 27 with non-CTD-ILD, with 68 healthy subjects as the control group. According to European Respiratory Association Guidelines, alveolar nitric oxide (CaNO) concentration and fractional exhaled nitric oxide (FeNO) level were measured at the flow rates of 50 and 200 mL/s. The predictive level of CaNO was analyzed using receiver-operating characteristic curve (ROC), and the correlations between CaNO and pulmonary function indicators were examined in the patients with ILD. CaNO, FeNO50, and FeNO200 levels were significantly higher in patients with ILD than in the healthy controls. Logistic regression analysis showed that lowered levels of CaNO and FeNO200 were risk factors for ILD. ROC curve analysis showed that the area under the curve (AUC) of CaNO combined with FeNO200 was 0.829 (95% CI: 0.752-0.906) for the diagnosis of ILD. In patients with ILD, CaNO levels were negatively correlated with DLCO%pred (r=-0.471, P < 0.05). Subgroup comparison showed a significantly higher CaNO level in CTD-ILD group than in non-CTD-ILD group. The AUC for CaNO to discriminate CTD-ILD from non-CTD-ILD was 0.725 (95% CI: 0.576 to 0.875). CaNO has a potential value in the diagnosis of ILD and differential diagnosis of CTD-ILD.

Introduction: The high resolution computed tomography (HRCT) is an important part in the diagnostic approach of interstitial lung disease (ILD) associated with connective tissue diseases (CTD) by providing detailed information on the elementary lesion and the radiological pattern of ILD. Aim: to point out the role of HRCT in the diagnosis of ILD associated with CTD (ILD-CTD). Methods: A Retrospective descriptive study was conducted between 2008 and 2017. Data of 24 patients presenting ILD-CTD were collected. A review of HRCT was performed by a radiologist without knowledge of the CTD. Results: Predominant elementary lesion of ILD associated with dermatomyositis (9 cases) was ground glass opacity (n = 9) followed by consolidation (n = 6). Non Specific Interstitial Pneumonia (NSIP) was the most reported pattern (5 cases). Ground glass opacity was also the predominant elementary lesion for the 2 cases of scleroderma and in Sj&#246gren’s syndrome (4 cases/5). NSIP was the predominant radiological presentation in these two CTD. Lymphoid interstitial pneumonia revealed Sj&#246gren’s syndrome in one case. In rheumatoid arthritis (6 cases), the elementary HRCT lesions were irregular interlobular septal thickening (n = 4) and honeycombing (n = 4) consistent with Usual Interstitial Pneumonia (UIP) in 2 cases. Similarly UIP has been described for the 2 patients with lupus and mixed connective tissue disease. Conclusion: HRCT plays an important role in the management of ILD-CTD. Description of the HRCT elementary lesions and the radiological pattern of ILD can be helpful for CTD’s diagnosis.

Deep learning for screening of interstitial lung disease patterns in high-resolution CT images